Kishimoto 2005 Wh-In-Situ and Movement in Sinhala Questions (MLLT 3)

Natural Language & Linguistic Theory (2005) 23: 1–51 Ó Springer 2005
HIDEKI KISHIMOTO
WH-IN-SITU AND MOVEMENT IN SINHALA QUESTIONSw
ABSTRACT. This article shows that Sinhala, a wh-in-situ language, implements

movement of a Q-element to determine the scope of wh-phrases; this movement,
which displays the behavior of a phrasal category, may be induced either in overt
syntax or in LF. Covert Q-movement observes island conditions in the same manner
as overt phrasal category movement. When the option of Q-movement is not
available, Sinhala makes use of a strategy to merge a null operator directly in its local
scope position to fix the scope of a wh-phrase.
1. INTRODUCTION
Sinhala (Indo-Aryan; Sri Lanka) is a wh-in-situ language1 in which

wh-questions have some notable properties. First, the scope of wh-
in-situ must be specified either by a Q-particle or by a special verbal
marking. Second, a Q-particle associated with wh-in-situ can appear
in various positions, either in a clause-final position where a
w
Earlier versions of the present paper were presented at a KLS workshop,
November 1994, at the 112th meeting of the Linguistic Society of Japan, June 1996,
and Linguistics Colloquium at Cornell University, October 1998. I am grateful to
James Gair, John Whitman, Paul Hagstrom, Chris Collins, Masayoshi Shibatani,
Yoshihiro Nishimitsu, Hiroshi Mito, Yoshiki Ogawa, Akira Otani, Yoshihisa Fujii,
Shuichi Yatabe, Masaki Sano, Yoshie Yamamori, and anonymous reviewers for
comments and suggestions. I am also thankful to Carol Georgopoulos and Joan
Maling for assistance on both the content and the form of this paper. I am deeply
indebted to Dileep Chandralal for acting as a primary informant. Without his
assistance and patience, this paper would never come into existence. Other infor-
mants consulted include Ketipearachchi Ysaranta, Kalyani Weerasinghe Ketipe-
arachchi, and Milton Rajaratne.
1
Gair and Sumangala (1992, p. 93) suggest that Sinhala could be a ‘move-wh’
language, basing their discussion on examples like (i) (adapted from their (3a) and
(3c)):
ðiÞa: amma ½Siri ti kieuwa kiy la kalp naa keruwe mok-ak d i :
e e e
mother Siri read-A that thinking did-E what Q
What was it that mother thought that Siri read?
2 HIDEKI KISHIMOTO
wh-phrase acquires scope, or in a constituent position typically

adjacent to a wh-phrase, where it marks a focused wh-constituent. In
this paper, I argue that these properties of Sinhala wh-questions
follow naturally from the assumption that a Q-particle is first merged
in a position to mark a wh-constituent and is then moved to clause-
final position as an operator encoding the scope of wh-in-situ. This
movement occurs either in overt syntax or in LF.
A major objective of the present paper is to show that it is Q-
movement, rather than movement of a wh-phrase, which is used to
form an operator-variable structure in a wh-question, and that a Q-
element, while delimiting a wh-constituent in its Merge position,
serves as an operator that assigns scope to its host wh-in-situ. I argue
that irrespective of whether the Q-element appears in Merge position
or in scope position in overt syntax, it eventually ends up in the scope
position, where it encodes the scope of wh-in-situ.
Significantly, even if Q-movement is induced in LF, it observes
island constraints in the same way as overt phrasal movement; Q-
particles behave like phrasal elements in this respect. It is often ob-
served that island effects do not emerge in wh-in-situ languages, even
when wh-phrases are deeply embedded in islands. I argue that this
stems from the fact that wh-phrases serve as bound variables that do
not undergo movement. However, island effects are incurred if a Q-
element, which is used to fix the scope of a wh-phrase, is merged in a
Footnote 1 (Continued)
ðiÞb: mok-ak d i amma ½Siri ti kieuwa kiy la kalp naa keruwe?
e e e
what Q mother Siri read-A that thinking did-E
What did mother think that Siri read?
However, the wh-words in (i) are not displaced by wh-movement but by other syn-
tactic operations (i.e., pseudo-clefting in (ia) and scrambling in (ib)), since constit-
uents other than wh-words can be moved:
ðiiÞa: amma ½Siri ti kieuwa kiy la kalp naa keruwe pot i ?
e e e
mother Siri read-A that thinking did-E book
It was the book that mother thought that Siri read.
b: pot i amma ½Siri ti keruwa kiy la kalp naa keruwa:

e e e
book mother Siri read-A that thinking did-A
The book, mother thought that Siri read.
Since Sinhala does not have a syntactic operation of wh-movement (in any strict
sense), it is reasonable to say that Sinhala is a ‘wh-in-situ’ language.
WH-IN-SITU IN SINHALA QUESTIONS 3
position where the later movement violates island conditions. When

the Q-element is merged outside islands, island violations do not occur
irrespective of whether movement takes place in overt syntax or in LF.
Some lexically-specified wh-phrases are not associated with a
detachable Q-particle, and display idiosyncratic behavior with regard to
their scope, in that they can only take scope over the local clause in which
they reside. I argue that this follows from the fact that the assignment of
scope to these wh-phrases can only be achieved with recourse to the
strategy of base-generating a null operator in the closest scope position,
since such wh-phrases are unable to make use of movement of a Q-
element to form an appropriate LF structure. The discussion leads to the
conclusion that wh-phrases are always construed as bound variables
(rather than wh-operators), whose scope must be interpreted by way of
an external scope assigner, i.e., a Q-particle or a null operator.
The discussion proceeds as follows. In section 2, I summarize the
basic facts of Sinhala wh-questions, and argue that a Q-element,
which can be construed as a non-projected head, behaves like a
phrasal element in the syntax. Section 3 illustrates that Q-movement,
which assigns scope to wh-in-situ, takes place in LF if it is not dis-
placed in overt syntax. Section 4 shows that when the option of
raising a Q-element is not available to generate the scope of wh-
phrases, Sinhala resorts to the strategy of base-generating a null
operator in the local scope position for scope assignment. Conclu-
sions are presented in section 5.
2. WH-CONSTITUENTS IN SINHALA
2.1. Particle-Predicate Concord

A Sinhala wh-interrogative is typically formed with a Q-element
occurring contiguous with a wh-element:
ð1Þ Chitra mon wa d gatte?

e e
Chitra what Q bought-E
What did Chitra buy?
One characteristic property of questions in which the Q-element d e

occurs immediately after the host wh-phrase is that the verb bears a
special ending, i.e., the -e ending which is glossed as ‘-E’. This verbal
ending differs from the neutral -a ending shown in (2) and glossed as
‘-A’:
4 HIDEKI KISHIMOTO
ð2Þ Chitra pot gatta:

e
Chitra book bought-A
Chitra bought the book.
In a matrix wh-question where the Q-particle d stands next to a wh-

e
in-situ, the verb cannot have the ordinary -a ending, as illustrated in
(3):

ð3Þ Chitra mon wa d gatta?
e e
Chitra what Q bought-A
In a wh-in-situ language like Japanese, a question particle occurs only

in clause-final position; in Sinhala, d generally cannot be placed
e
clause-finally in direct wh-questions:

ð4Þ Chitra mon wa gatta=gatte
e d ? e
Chitra what bought-A=bought-E Q
(4) is ill-formed regardless of whether the verb receives the -e marking

or the -a marking.
Wh-words in Sinhala are indefinite pronouns whose interpretation
depends on the kind of Q-element. For example, when mon wa ‘what,
e
thing’ takes the affix -t, as in mon wa-t, it means ‘anything’, but if it
e
takes -hari, as in mon wa-hari, it means ‘something’. Owing to this
e
property of wh-words, the presence of d is mandatory in an ordinary
e
wh-question; if no Q-element is present, the sentence is judged ill-
formed regardless of whether the verb bears the -a or -e ending, as
illustrated in (5):

ð5Þ Chitra mon wa gatte=gatta?
e
Chitra what bought-E=bought-A
The -e marking must co-occur with the Q particle d located adjacent

e
to a wh-phrase; in this paper, this co-occurrence requirement is
referred to as PARTICLE-PREDICATE CONCORD.
In wh-questions, the -e ending encodes the scope of wh-elements.
The examples in (6) illustrate that in a wh-question where d is e
adjacent to a wh-phrase, the scope of the wh-phrase is interpreted
relative to this verbal marking:
ð6Þa: Ranjit ½kau d aawa ekiy la danne? e

Ranjit who Q came-A that know-E
Who does Ranjit know came?
b: Ranjit ½kau d aawe e kiy la dann wa:e e
Ranjit who Q came-E that know-A
Ranjit knows who came.
In (6a), the matrix verb danne ‘know’ bears the -e marking, and the
sentence is understood to be a matrix wh-question; in (6b), the lower
verb aawe ‘came’ bears the -e marking, and the sentence is under-
stood as an embedded wh-question. The assumption that the -e
marking serves to encode the scope of a wh-phrase gains additional
plausibility from (7):

ð7Þ kau d ½Ranjit aawe
e kiy la dann wa: e e
who Q Ranjit came-E that know-A
In (7), kau d ‘who Q’ is located in the matrix clause but the

e
embedded verb receives the -e ending. This type of wh-scope,
involving a lowering operation, is illegitimate cross-linguistically, and
the ungrammaticality of (7) confirms that the scope of a wh-word is
specified by the special -e marking on the verb.
Note that the -e and -a affixes can only appear on the finite form of
a verb, and are incompatible with other verbal forms such as parti-
ciples, infinitivals and adnominals. Hence, the scope of a wh-phrase
can only be marked on a finite verb, which may occur either in the
matrix clause or in the embedded clauses selected by complementizers
like kiy la ‘that’, which introduces an ordinary subordinate clause,
e
and ki n ‘that’, which introduces a noun complement clause.
e e
Notably, the verb which immediately precedes a relative clause head
takes an adnominal form with the - ending, which differs from the -e
e
or -a ending. This indicates that the verb taking an adnominal form is
not capable of marking the scope of a wh-phrase.2 Nor is it possible
for a Q-element to occur at the right end of the adnominal verb; wh-
phrases can never take scope over a relative clause.
2
The -e marking is not used in relative clause formation even though relativiza-
tion involves some kind of operator binding. The reason is that the special -e affix is
usable only in focus constructions where focus is syntactically separated from pre-
supposition. Since relativization does not fulfill this function, the -e ending does not
appear on the verb in the relative clause.
6 HIDEKI KISHIMOTO
A Q-particle d associated with a wh-phrase may occur in clause-

e
final position, although the contexts in which it can appear there are
fairly restricted (see Gair 1983; Kishimoto 1992; Gair and Sumangala
1992). For example, clause-final Q-placement is possible when a wh-
phrase is embedded within the complement clause of a verb like
dann wa ‘know’, and also in direct wh-questions formed with kiidenek
e
‘how many (animate)’ and kii-ak ‘how many (inanimate)’:
ð8Þa: Ranjit ½kauru aawa d kiy la dann wa:3
e e e
Ranjit who came-A Q that know-A
b: kiidenek pot kieuwa d ?
e e
how:many book read-A Q
How many (people) read the book?
With the wh-words kiidenek and kii-ak, d may be placed clause-
e
finally even in the matrix clause. In (8a) and (8b), where d occurs in e
clause-final position, the verb takes the ordinary declarative -a end-
ing, not the -e ending.
The clause-final Q-particle does not mark a wh-constituent on the
surface; instead, it specifies wh-scope, behaving in a way similar to the
-e ending on a predicate in a wh-question where d is attached to a e
wh-constituent:
ð9Þ a: Ranjit ½kiidenek en wa kiy la dann wa d ?
e e e e
Ranjit how:many come-A that know-A Q
How many (people) does Ranjit know will come?
b: Ranjit ½kiidenek en wa d kiy la dann wa:
e e e e
Ranjit how:many come-A Q that know-A
Ranjit knows how many (people) will come.
Example, (9a) is only understood as a matrix wh-question, and (9b) as
an embedded wh-question; this shows that the clause-final Q-element
marks the scope of a wh-phrase. This view gains support from the
unacceptability of (10):

ð10Þ kiidenek ½Ranjit en wa d kiy la dann wa:
e e e e
how:many Ranjit come-A Q that know-A
3
The wh-word meaning ‘who’ is kauru, but when d is adjacent to the wh-word,
e
-ru is dropped, as in kau d .
e
In (10), d appears in the embedded clause, while the host wh-phrase

e
is in the matrix clause. Since (10) patterns with (7), it can be
reasonably concluded that the clause-final d assumes a scope- e
marking function.
Whenever it is possible for the Q-particle d to appear in clause-
e
final position, it is also possible for it to occur instead in a position
adjacent to its host wh-phrase; compare (11a,b) to (8a,b).
ð11Þ a: Ranjit ½kau d aawe ekiy la dann wa:

e e
Ranjit who Q came-E that know-A
b: kiidenek d pot kieuwe?
e e
how:many Q book read-E
If the Q-element is attached to a wh-phrase, then the -e ending

appears on the verb in the clause where the wh-word takes scope, as
illustrated by the minimal pairs in (8) and (11).4 Not surprisingly, in
cases in which two possible sites for Q-elements are available, a single
4
The -e marking signaling the scope of a wh-phrase can only appear in the clause
where the host wh-phrase takes scope. If this verbal marking occurs in other places,
the sentence is unacceptable:
ðiÞa: Ranjit ½Chitra kiidenek d daekka= daekke kiy la danne?

e e
Ranjit Chitra how:many Q saw-A=saw-E that know-E
How many (people) does Ranjit know that Chitra saw?
b: Ranjit ½Chitra kiidenek d daekke kiy la dann wa= danne:

e e e
Ranjit Chitra how:many Q saw-E that know-A=know-E
Ranjit knows how many (people) Chitra saw.
In a matrix wh-question like (ia), the -e marking can appear only on the matrix verb,
but not on the embedded verb. In an embedded wh-question like (ib), only the
embedded verb can have the -e marking. In a wh-question in which d occurs in e
clause-final scope position, the -e marking cannot appear anywhere:
ðiiÞ a: Ranjit ½Chitra kiidenek daekka= daekke kiy la dann wa d ?

e e e
Ranjit Chitra how:many saw-A=saw-E that know-A Q
How many (people) does Ranjit know that Chitra saw?
b: Ranjit ½Chitra kiidenek daekka d kiy la dann wa= danne:

e e e
Ranjit Chitra how:many saw-A Q that know-A=know-E
Ranjit knows how many (people) Chitra saw.
8 HIDEKI KISHIMOTO
wh-phrase may not be associated with two instances of the Q-particle

d ;5 as illustrated in (12):
e
ð12Þa: Ranjit ½kau d aawe=aawa

e d kiy la dann wa: e e e
Ranjit who Q came-E=came-A Q that know-A
b: kiidenek d pot kieuwe=kieuwa d ?
e e e
how:many Q book read-E=read-A Q
Verbs permitting the clause-final d in their complement clause, as
e
in (8a),
R include dann wa ‘know’, hoya b r n wa ‘examine’,
e e ee
parik aa k r n wa ‘look into, inspect’, and teeren wa ‘understand’.
e ee e
However, there are other verbs which do not allow d to be placed in e
clause-final scope position, as shown in (13):
ð13Þ a: Ranjit ½kauru aawa d kiy la aehuwa:

e e
Ranjit who came-A Q that asked-A
Ranjit asked who came.
b: Ranjit ½kau d aawe ekiy la aehuwa: e
Ranjit who Q came-E that asked-A
Ranjit asked who came.
This class of verbs includes ahan wa ‘ask’, prasn k r n wa ‘ques-
e e ee e
tion’, and hiten wa ‘consider’. The class of verbs which allow the
e
placement of d in clause-final scope position cannot simply be
e
specified by the ability to take an interrogative subordinate clause,
Footnote 4 (Continued)
Further, as shown in (iii), a non-interrogative embedded clause can have neither the
-e marking nor the clause-final Q-particle:
ðiiiÞa: Ranjit ½Chitra kiidenek d daekke kiy la wiswaas k r n wa=kendiruwa:

e e e ee e
Ranjit Chitra how:many Q saw-E that believe-A=whispered-A
Ranjit believes/whispered how many (people) Chitra saw.
b: Ranjit ½Chitra kiidenek daekka d kiy la wiswaas k r n wa=kendiruwa:

e e e ee e
Ranjit Chitra how:many saw-A Q that believe-A=whispered-A
Ranjit believes/whispered how many (people) Chitra saw.
In wh-questions, only when d remains in its ‘delimiter’ position can the -e affix occur
e
on the verb in the clause over which the associated wh-word takes scope.
5
Similar facts obtain for other Q-particles as well, since a wh-phrase cannot be
bound by more than one Q-element.
because ahan wa ‘ask’ does not allow d to be placed at the end of its
e e
subordinate clause; nor can verbs taking the clause-final d be spec-
e
ified by their ability to select a declarative complement in addition to
an interrogative complement:
ð14Þ Ranjit ½Chitra aawa kiy la dann wa=hiten wa=

e e e
Ranjit Chitra came that know-A=consider=
*hoya b r n wa/*ahan wa.
e ee e
examine-A/ask-A
Ranjit knows/considers/examines/asks that Chitra came.
As shown in (14), there is no correlation between the class of predi-

cates permitting the clause-final Q-placement in the complement
clause and the class of predicates selecting a non-interrogative com-
plement. For present purposes, it is sufficient to note that the
sequence ‘Verb+d kiy la’ is allowed only for a proper subset of
e e
verbs that can take an interrogative complement (see Gair 1983).
In cases where there are two options for placement of the Q-par-
ticle, either attached to the host wh-phrase or in clause-final position,
these two options are used in different discourse contexts. The version
with the particle attached to the wh-phrase is uttered when the speaker
assumes that there is at least one value which satisfies the proposition,
whereas clause-final particle placement indicates no such presuppo-
sition. In other words, a clause-final d is used in a discourse context
e
where the set of individuals that can fill the value of the wh-word might
be empty. The speaker therefore would not be surprised to receive the
answer (15b) in response to (15a):
ð15Þ a: Q : kiidenek pot kieuwa d ?

e e
how:many book read-A Q
b: A : kauru-wat kieuwe naeae:
anyone read not-A
No one read it.
By contrast, since wh-attached d is uttered in a context in which the

e
speaker anticipates that there is at least one value satisfying
the proposition, the speaker would not expect the answer (16b) to the
question (16a):
10 HIDEKI KISHIMOTO
ð16Þ a: Q: kiidenek d pot kieuwe?

e e
how:many Q book read-E
How many (people) was it that read the book?
b: A : #kauru-wat kieuwe naeae:
anyone read not-A
No one read it.
Essentially the same situation holds for (8a) and (11a). In uttering a
sentence of the type (8a), the speaker holds that the set of individuals
which satisfy the proposition might be empty. Hence, the following
question-answer pair is natural:6
ð17Þa: Q: oyaa ½ kauru aawa d kiy la dann wa d ?
e e e e
you who came-A Q that know-A Q
Do you know who came?
b: A: oo: kauru-wat aawe naeae:
yes anyone came not-A
Yes. No one came.
On the other hand, a sentence like (11a) is uttered in a discourse

context where the speaker presumes that there should exist at least
one individual who came. Thus, the answer in (18b) strikes the
speaker as unexpected:
ð18Þa: Q: oyaa ½kau d aawe kiy la dann wa d ?
e e e e
you who Q came-E that know-A Q
Do you know (who it was) who came?
b: A: #oo: kauru-wat aawe naeae:
yes anyone came not-A
Yes. No one came.
When a wh-question involving an ordinary wh-phrase like kauru

‘who’ is directly addressed to the hearer as a matrix question, the
speaker seeks the value of the wh-word, while presupposing that the
set of individuals satisfying the proposition is not empty. Thus, in
6
In (17a), the semantic interpretation is not equivalent to ‘Do you know whether
anyone came?’, since the embedded clause is a wh-interrogative, which asks for the
identity of individuals, although the speaker anticipates that the actual value that fills
the wh-word may be null. Needless to say, in this case, the embedded d cannot be
e
replaced by d -naedd ‘whether,’ without changing the meaning.
e e
ordinary direct wh-questions, d cannot be placed clause-finally.7 The

e
placement of d in clause-final scope position can be effected when the
e
set of individuals satisfying the proposition can be assumed to be
possibly empty.
With regard to scope, the Q-particle d has a function similar to
e
the focus particles tamay and y and the negative focus particle newey,
all of which can be used to mark focused constituents, as illustrated
by the example in (19):
ð19Þ Chitra ee pot tamay kieuwe:

e
Chitra that book FOC read-E
It was that book that Chitra read.
In (19), the particle tamay specifies a focused constituent, and the -e

marking specifies the scope of the delimited constituent.
It is worth noting that focus particles generally are allowed to
appear in clause-final position as well, as illustrated in (20):
ð20Þ Ranjit ee pot kieuwa tamay:

e
Ranjit that book read-A FOC
Certainly, Ranjit read that book.
When tamay occurs clause-finally, the verb takes the ordinary -a

ending, rather than the -e ending. Clause-final tamay does not delimit
a focused constituent, and focus may fall on any constituent in the
clause. Thus, (20) can mean ‘It was Ranjit who read that book’, ‘It
was that book that Ranjit read’ or ‘Ranjit did read that book’.
When not associated with a wh-phrase, the particle d can be usede
to indicate a yes/no question. This yes/no particle can be placed in
focus-marking position or clause-final scope position, just like tamay:
ð21Þa: Chitra ee pot d kieuwe?e e

Chitra that book Q read-E
Was it that book that Chitra read?
b: Chitra ee pot kieuwa d ?
e e
Chitra that book read-A Q
Did Chitra read that book?
7
Essentially the same account can be carried over to the ill-formedness of (13a),
where d occurs in clause-final position.
e
12 HIDEKI KISHIMOTO
Unlike the wh-question particle d , the yes/no question particle may

e
be placed freely in clause-final position even in the matrix clause.
When it is contiguous with a nominal constituent, like (21a), it marks
a focused constituent, but when it occurs at the end of a clause, like
(21b), it simply indicates that the relevant clause is a yes/no inter-
rogative.8
Note further that yes/no interrogation may be indicated by
d -naedd ‘whether’.9 This particle differs, however, from the yes/no
e e
question particle d , in that it can only be placed in clause-final
e
position:
ð22Þa: Chitra ee pot kieuwa d -naedd ?
e e e
Chitra that book read-A whether

b: Chitra ee pot d -naedd e e e kieuwe=kieuwa?
Chitra that book whether read-E=read-A
I assume that d -naedd is a particle that occupies a clause-final
e e
complementizer position, just like a clause-final question particle d . e
8
The yes/no question particle, which can be construed as having the same
function as focus particles like tamay, behaves differently from the wh-question
particle, in that it is not possible to have multiple foci in a single clause:

ðiÞ Chitra d ee pot d kieuwe?
e e e
Chitra Q that book Q read-E
The unacceptability of (i) parallels the unacceptability of (ii), which involves multiple
foci with the focus particle tamay:

ðiiÞ Chitra tamay ee pot tamay kieuwe:
e
Chitra FOC that book FOC read-E
Certainly, Chitra read that book.
By contrast, the wh-question particle allows for multiple foci:
ðiiiÞ kau d mon pot d kieuwe?

e e e e
who Q what book Q read-E
Who read what book?
Thus, the yes/no question particle patterns with the focus particle tamay, but not
with the wh-question particle.
9
To be more precise, the complementizer d -naedd can be glossed as ‘whether-or-
e e
not’, or ‘Q-not-Q’.
A clause introduced by d -naedd corresponds in form to a whether-

e e
clause in English, but unlike a whether-clause in English, it can serve
as a direct yes/no question.
Returning to the discussion of wh-questions, the Q-element d e
most typically appears immediately after its host wh-word. However,
it is not directly attached to a lexical head, but rather is attached to
some maximal projection. In (23a) and (23b), d appears adjoined to
e
the DP containing the wh-phrase, and in (23c), it is preceded by the
postposition ekk ‘with’:
e
ð23Þ a: Chitra ½mon pot d gatte?

e e e
Chitra what book Q bought-E
What book did Chitra buy?
b: Chitra ½kaa-ge amma d daekke? e
Chitra who-GEN mother Q saw-E
Whose mother did Chitra see?
c: Chitra ½kauru ekk d kataa k lee?
e e e
Chitra who with Q talk did-E
With whom did Chitra talk?
It is not possible to place d immediately after a wh-word embedded

e
inside PP or DP, since DPs and PPs constitute islands in Sinhala (see
section 3).
In (8a), the Q-element d appears in the scope position, which lies
e
between the complementizer kiy la and the finite verb in the
e
embedded clause. The same Q-element can also occur to the right of
the complementizer as an additional site for Q-placement, in which
case d does not mark the scope of wh-in-situ:
e
ð24Þ Ranjit ½kauru aawa kiy la d danne= dann wa?

e e e
Ranjit who came that Q know-E=know-A
Who does Ranjit know came?
In (24), since an independent scope marking, namely the -e marking

on the matrix verb, must be present, we can reasonably assume that
the Q-element d is used as a delimiter marking the delimited con-
e
stituent (see section 2.2).
In conclusion, Sinhala wh-questions are typically formed with d e
appearing either (i) attached to the right edge of the maximal pro-
jection including a wh-word in it, as in (23a–c), or (ii) to the imme-
diate right of the kiy la-complementizer, as in (24). In these two
e
14 HIDEKI KISHIMOTO
cases, d delimits a focused wh-constituent, and the scope of the wh-

e
element is specified by the special -e ending on the verb, producing
particle-predicate concord. There are also contexts in which d may e
be placed in scope position, that is, (iii) to the right of the matrix verb,
as in (9a), or (iv) between the finite verb and the complementizer in
the embedded clause, as in (9b). In the latter two cases, d marks the
e
scope of wh-in-situ, and the verb has an ordinary ending.
2.2. The Status of Q-Particles

Having surveyed the basic properties of Sinhala wh-questions, we are
now in a position to discuss some of the theoretical assumptions
pertaining to the nature of Q-particles. In this section, I show that
while the Q-particle d is merged in a delimiter position to mark a
e
delimited wh-constituent, it serves as an operator to bind wh-in-situ
after it is moved to its scope position.
Let us begin by noting that a Q-particle used to delimit a wh-
constituent can be attached to a number of different types of con-
stituents without changing their categorical properties; this suggests
that the operation to generate d involves adjunction (see Aoyagi
e
1999). In addition, since d is a particle that does not project any
e
further, we can assume that it has the properties of a non-projected
head in the sense of Chomsky (1995b). A non-projected head can, in
principle, be either a maximal or a minimal projection. But since d e
must be adjoined to a maximal projection when it is merged, as noted
earlier, it is plausible to hypothesize that d counts as a phrasal ele-
e
ment in the syntax. If so, we can assume that the wh-expression
mon wa d ‘what Q’ should have the following structure:
e e
ð25Þ DP
DP d
e
mon wa
e
When a Q-particle is merged in its Merge position, it is merged (as an

XP element) with a maximal projection which includes the wh-form.
This is a reasonable assumption, since only XP can adjoin to XP
(Chomsky 1995a). I also assume that the relevant Merge operation
forms a two-segment category (see Chomsky 2000).
The next question is where the clause-final d , which encodes the
e
scope of wh-in-situ, is located in the clause structure. To see this, first
note the general fact that in Sinhala, the complementizer kiy la ‘that’
e
occurs in both declarative and interrogative clauses, and that a

clause-final d is allowed to co-exist with kiy la, as in (8a) and (9b)
e e
above. That d , which I assume is an operator binding a wh-phrase,
e
can appear to the immediate left of the complementizer suggests that
CP recursion is allowed in Sinhala. I assume then that Sinhala has
two layers of CP projections, as represented in (26):
CP1
ð26Þ
CP 2 kiy la
e
C 2' d e
TP C2
The lower CP, filled by an operator such as the Q-element as in (26),

is referred to as CP2 , so as to distinguish it from the upper CP which
is occupied by the ordinary complementizer CP1. This analysis is in
fact in accord with the basic claim shared by researchers advocating
CP stacking (see Browning 1996, and also Culicover 1992; Authier
1992; Bhatt and Yoon 1991; Whitman 1989; Kim 1989).
Since the clause-final d , which assigns scope to a wh-phrase, is a
e
non-projected head which acts like a phrasal element in the syntax
(by hypothesis), I assume that d occupies [Spec, C2 ], rather than the
e
C2 head position. This assumption receives support from the fact that
the clause-final Q-element d counts as an operator that binds and
e
forms an A-bar chain with its associated wh-phrase. The wh-expres-
sion bound by d must be a maximal projection of a category, but not
e
just a head or an X0 category, for the theta criterion to be satisfied. If
so, the Q-element d must also have the status of XP to meet the
e
requirement of ‘chain uniformity’ (see Chomsky 1995b, 2000).
The Q-element located in scope position does not mark a delimited
wh-constituent. But as discussed below, a delimited wh-constituent
must be picked out by the Q-element in a wh-question. This suggests
that the Q-element can be placed in clause-final position only by means
of movement, which moves it to the scope position from a constituent
position where it is merged to mark a delimited wh-constituent, and
that the Q-element cannot be based-generated in its scope position as a
complementizer. Now, given the standard assumption that XP moves
into Spec, and does not target a head position, we can maintain the
hypothesis that the clause-final Q-element is a phrasal element that
16 HIDEKI KISHIMOTO
occupies [Spec, C2 ].10 In Sinhala, the operator position [Spec, C2 ] must

stand at the right end of a clause, since a clause-final Q-element is
located between the finite verb and the complementizer in the
embedded clause, and to the right of the finite verb in the main clause.
Note further that the Q-particle d is construed as a dependent
e
element affixing to whatever constituent appears to its left. This
property of d can also be naturally explained on the current
e
assumptions. Since a Q-particle is a non-projected element, which
may count as a minimal projection (as well as a maximal projection),
we can assume that in PF it does not have the status of a phrase but
functions like an X0 head. Given the assumption that a Q-particle
acts like a head element at PF, it is easy to see that d is affixed to a
e
constituent on its left, behaving like a clitic.11
Let us now proceed to discuss the question of how d is placed in e
various positions, and in particular, how it can occur in either ‘scope’
position or ‘delimiter’ position. I propose that d can be merged (i.e.,
e
base-generated) in any delimiter position where the constituent it
delimits can be turned into a simple wh-expression. I argue that since
clause-final d does not delimit a wh-constituent, d can only be placed
e e
in clause-final scope position via movement from a delimiter position.
The Q-particle d serves primarily to delimit a wh-constituent in a
e
wh-question, in a way similar to focus particles like tamay. As for its
possible Merge site, then, it is reasonable to hypothesize that d can e
only be merged in a ‘delimiter’ position which marks a delimited wh-
constituent. On this view, if d is to mark wh-scope (as an operator to
e
form an A-bar chain with the wh-phrase), it must be moved to the
scope position, as shown in (27b):
ð27Þ a: ½CP2 ½TP . . . WH d . . . e
b: ½CP2 ½TP . . . WH ti . . .d i e
I propose that d can potentially be merged in any ‘delimiter’ position

e
where it picks out a wh-constituent. Thus, although d is most typi- e
cally merged to the closest possible maximal projection to the host
wh-word, it may also be merged in a distant position separate from
the wh-word, as long as it can c-command and bind the wh-word.
Since d in delimiter position serves to pick out a focused constituent
e
10
Likewise, when a focus particle like tamay appears in clause-final position, I
assume that it appears in [Spec, C2 ].
11
Since focus particles are allowed to occur in the same position as the Q-particle,
I assume that in the sequence ee pot tamay ‘that book FOC’, tamay is adjoined to
e
the DP ee pot .
e
in a wh-question, I assume that its merger is licensed when a focus

feature [+F] on the wh-word is legitimately percolated up to the
delimited wh-constituent.
One might argue, however, that d can be merged even more
e
freely, and that Merge can target both scope-marking and delimiter
positions. If this is the case, the option where d is directly merged in
e
the scope position, should also be available, as represented by (28):
ð28Þ ½CP2 ½TP . . . WH . . .d
e
In Sinhala, however, there is good reason to believe that d , which is

e
associated with a wh-phrase, cannot be directly merged in its scope-
marking position.
Empirical evidence in favor of the view that the wh-question
particle d may not be base-generated in scope position may be
e
obtained by looking at a ‘minimal’ answer to a wh-question. To see
this, consider (29):
ð29Þ a: Q: Chitra kiidenek d daekke?
e
Chitra how:many Q saw-E
How many (people) did Chitra see?
b: A: tundenek:
two
Two (people).
The wh-question in (29a) can be answered in a number of ways, but
a minimal answer can be made by supplying only the value of the
wh-phrase, as in (29b). Interestingly, in a case in which a Q-element
is merged in a position separate from its host wh-phrase, it is not
possible to supply just the value of the wh-word as a minimal
answer:
ð30Þa: Q: Chitra ½kiidenek ekk d sellan k lee?
e e e
Chitra how:many with Q play did-E
With how many (people) did Chitra play?
b: A: ? tundenek:
two
Two (people).
c: A0: tundenek ekk : e
two with
With two (people).
18 HIDEKI KISHIMOTO
Example (30a) represents a case in which a wh-word is embedded in a

PP. Since the sequence ½kiidenek d ekk ‘how.many Q with’ is not
e e
possible, the Q-particle d must be placed outside the PP. For (30a), a
e
licit minimal answer is the one which repeats the PP along with the
value of the wh-word.
The same constraint is imposed on (31a), where a wh-word is
embedded inside a complex DP island:
ð31Þa: Q: Chitra ½ ½kiidenek aawa ki n ka kataaw d aehuwe?

e e e e e
Chitra how:many came that rumor Q heard-E
Chitra heard the rumor that how many (people) came?
b: A : ? tundenek:
two
Two (people).
c: A0 : ½tundenek aawa ki n ka kataaw :
e e e e
two came that rumor
The rumor that two (people) came.
In response to (31a), it is not legitimate to give just the value of the

wh-word. In order to give a well-formed minimal answer, it is nec-
essary to recapitulate the entire material marked with d , along with e
the value of the wh-word, as in (31c).12
12
As pointed out to me by Joan Maling (p.c.), English echo questions work
somewhat similarly, since (ic), but not (ib), is a felicitous answer to a question like
(ia):
ðiÞa: Q : John heard the rumor that WHO was coming?

b: A : ? Robin:
0
c: A : The rumor that Robin was coming:
The unacceptability of (ib) stands in contrast to the acceptability of (iib), which is

provided as an answer to (iia):
ðiiÞa: John heard that WHO was coming?

b: Robin:
In English, unlike Sinhala, no syntactic device of a delimiting marker is available, so

this fact might suggest that delimitation of a constituent is made relying on prag-
matics. In (ia), the complex DP, rather than the wh-phrase, is a dominant element in
the discourse, so that an answer like (ib), which supplies only the value of the wh-
phrase, is not appropriate.
In (29a), since d is directly attached to the wh-phrase, its minimal

e
answer is the one which only comprises the value of the wh-phrase.
But in (31a) d is directly merged to the DP containing a noun
e
complement, so its minimal answer must recapitulate the whole
complex DP together with the value of the wh-phrase. The important
point is that a licit minimal answer requires that the wh-constituent
delimited by d , to which I assume the [+F] feature of a wh-word is
e
percolated up, be recapitulated.
In cases in which the Q-element d appears in scope position, the
e
answer that recapitulates the entire sentence does not count as a
minimal answer. To answer the question in (32a), it is fully permis-
sible to provide just the value of the wh-word:
ð32Þa: Q : Chitra kiidenek daekka d ? e

Chitra how:many saw-A Q
How many (people) did Chitra see?
b: A : tundenek:
two
Two (people).
If d were directly merged in the scope position, the entire sentence

e
should be delimited as a wh-constituent. If so, its minimal answer
should be the one which spells out the entire sentence along with
the value of the wh-word, but clearly, this type of answer is not
minimal.
For a wh-question like (33a), where the wh-word is inside PP,
its minimal answer must repeat the PP, as shown in (33c), and
the answer just supplying the value of the wh-word does not count
as a minimal answer, as shown in (33b). Compare the examples in
(31).
ð33Þa: Q: Chitra ½kiidenek ekk sellan k laa d ?

e e e
Chitra how:many with play did-A Q
With how many (people) did Chitra play?
b: A : ? tundenek:
two
Two (people).
20 HIDEKI KISHIMOTO
c: A0: tundenek ekk : e

two with
With two (people).
Furthermore, in (34a), in which a wh-word is embedded in a noun
complement, just like (31a), the answer that recapitulates the complex
DP alongside the value of the wh-word is called for as its minimal
answer, even though d is placed at the end of the sentence:
e
ð34Þa: Q: Chitra ½ ½kiidenek aawa ki n ka kataaw

e e e e
Chitra how:many came-A that rumor
aehuwa d e
heard-A Q
Chitra heard the rumor that how many (people)
came?
b: A : ? tundenek:
two
Two (people).
c: A0: ½tundenek aawa ki n ka kataaw :
e e e e
two came-A that rumor
The rumor that two (people) came.
Thus, a comparison of the minimal answers to the questions above

shows that no change is observed when d occurs clause-finally. Since
e
these minimal answers reflect the constituent which d delimits, it e
must be the case that in (32a)–(34a), the Q-element picks out a
focused wh-constituent, located in a delimiter position (rather than at
the end of the sentence), in the same manner as the Q-element in
(29a)–(31a). Given this, it is easy to see that in (32a)–(34a), the Q-
element d is merged in a delimiter position where d occurs in (29a)–
e e
(31a), but is placed in the scope position on the surface as a conse-
quence of overt movement.13
It should be noted that when a wh-phrase is embedded in the
complement clause of a bridge verb like kien wa ‘say’, it is possible
e
for d to occur either contiguous with the wh-phrase or to the
e
immediate right of the complementizer. This suggests that d can be e
13
The type of question-answer pair found in (34) is often used to argue for the so-
called ‘pied piping’ analysis of wh-questions, which claims that a large-scale DP
containing a wh-phrase may be pied piped into [Spec, C] (see Nishigauchi 1990; Choe
1987; Pesetsky 1987, among others). The Sinhala facts indicate that what actually
undergoes movement is the Q-element, rather than a pied-piped constituent.
base-generated in two different ‘delimiter’ positions. (Recall that the

Q-element occurring to the right of a complementizer is in a delimiter
position, but not in a scope position.) Interestingly, possible minimal
answers differ depending on where the Q element occurs. First,
consider (35):
ð35Þa: Q: Chitra ½kau d aawa e kiy la kiiwe?
e
Chitra who Q came-A that said-E
Who did Chitra say came?
b: A: Ranjit:
In answer to a question like (35a), where d is contiguous with the
e
wh-phrase, it is fully permissible to provide just the value of the wh-
phrase as a minimal answer, as shown in (35b). But when the Q-
particle occurs to the immediate right of the complementizer, it is
necessary to repeat the embedded clause with the complementizer,
alongside the value of the wh-phrase, as illustrated in (36):
ð36Þa: Q: Chitra ½kauru aawa kiy la d kiiwe?
e e
Chitra who came-A that Q said-E
Who did Chitra say came?
b: A : ? Ranjit:
c: A0: Ranjit aawa kiy la: e

Ranjit came-A that
That Ranjit came.
The data indicate that in (36a), the whole embedded clause is

delimited as a wh-constituent. Note that if d were first merged in a
e
position adjacent to the wh-phrase and moved to the right of the
complementizer, the wh-phrase, rather than the entire complement
clause, would be delimited. In that case, we would expect that the
answer in (36b) should be licit, contrary to fact. In light of this
consideration, we can reasonably conclude that in (36a), d is directly
e
merged to the right of the complementizer.
When a wh-phrase is embedded in the complement clause of a
bridge verb, two different Merge sites for d are available. This
e
immediately raises the question of where an overtly moved d is e
merged in such a context. When d is positioned in the scope position,
e
as in (37a), it is possible to provide just the value of the wh-phrase as a
minimal answer:
22 HIDEKI KISHIMOTO
ð37Þa: Q: Chitra ½kiidenek aawa kiy la kiiwa d ?e e

Chitra how:many came-A that said-A Q
How many (people) did Chitra say came?
b: A: tundenek:
two
Two (people).
The fact that the wh-phrase is a delimited constituent in (37a) indi-
cates that d is first merged in a position contiguous with the wh-
e
phrase, rather than to the right of the complementizer, and then is
overtly moved to its scope position.
Now, if the clause-final placement of d involves operator move-
e
ment, the question to be addressed is what motivates it. Here, I
suggest that a formal feature which contributes to the morphological
realization of the -e marking on a finite verb is responsible for the
attraction of d . I argue that particle-predicate concord in Sinhala
e
may be adequately characterized if we assume that the formal feature
[+Q] on the verb needs to be checked off by d , which assigns scope
e
to its host wh-phrase, or to be more precise, by an ‘operator’ feature
in it (cf. Chomsky 1993, pp. 31–32). I propose that the morphological
realization of verb forms is dictated by the following morphological
rule:
ð38Þ A finite verb bears the -e ending if the feature [+Q] is present
on the verb at PF; otherwise, it bears the -a ending.
There is a sense in which the [+Q] feature should be checked and

deleted by d when it is attracted to [Spec, C2 ], since the special -e
e
marking can appear only on the verb in the clause where the wh-
phrase actually takes scope in wh-questions. (When d is not overtly
e
moved, [+Q] marks wh-scope, but in the present analysis, [+Q] is not
an operator that forms an A-bar chain with a wh-phrase.)
The overt movement of d must be feature-driven. If [+Q] assigned
e
to a predicate is weak, then overt movement of d is not required. In
e
that case, d remains in the original position where it is merged in
e
overt syntax.
ð39Þ ½CP2 ½TP . . . WH d . . . ½þQ

e
In (39), feature checking does not take place overtly, so that [+Q]
feeds into PF, and the verb receives the -e ending. (In this case, the
Q-particle is moved covertly to its scope position, which means that it

comes to function as an operator to form an A-bar chain with the
wh-phrase in LF; I will turn to this issue in section 3.) On the other
hand, if [+Q] is strong, then d , which is merged in a position to
e
delimit a wh-constituent, is overtly moved into its scope position for
feature checking.
ð40Þ ½CP2 ½TP . . . WH ti . . . d i

e
When d occurs in a position to mark scope, as in (40), the strong

e
feature [+Q] on the verb is checked and deleted by d in the checking
e
domain of C2 in overt syntax, so the verb yields the ordinary -a
ending. In this case, d , located in clause-final position, is morpho-
e
logically affixed to the verb, and marks the scope of the wh-word. I
assume here that in Sinhala the verb is head-raised from V to C2 and
resides in C2 in overt syntax, and that feature checking is executed
under a Spec-head configuration.
The assignment of a strong or weak feature is conditioned by a
semantic factor. The foregoing discussion on the position of Q-
particle indicates that a weak feature [+Q] is assigned where an
‘existential’ presupposition obtains, while a strong feature [+Q] is
assigned where no such presupposition is required. In the context
in which Q-movement is optional, the conditions for the assign-
ment of the strong [+Q] feature as well as the weak [+Q] feature
are met. Note that the semantic difference, which governs the overt
position of d , is not structurally represented in LF, but the cor-
e
rect interpretation still obtains, since the presupposition is encoded
as a semantic feature of the wh-word, which survives to the LF
output.
Overt movement of d displays the properties of phrasal A-bar
e
movement, just as in ordinary operator movement, in that movement
targets a scope position, i.e., CP2 (to form an operator-variable
structure with a wh-word), its dependency can be long distance, etc.
This state of affairs is naturally anticipated under the proposed
analysis, according to which a Q-particle behaves like a phrasal
category for the purpose of movement.
The next question to be raised is why the wh-question particle d e
is first merged to a constituent inside a clause, serving as a delimiter,
even when it appears in its scope position. The key to the answer
lies in the fact that the Q-element serving as a scope marker
appears in such a position that it cannot separate a legitimate
24 HIDEKI KISHIMOTO
wh-constituent from other constituents. In Sinhala wh-questions,

although d is sometimes merged to a maximal projection other
e
than its associated wh-phrase, the wh-constituent picked out by d e
must be one that can be wh-questioned. Thus, no matter how far d e
is merged away from the host wh-phrase, it can only be merged in a
position where the delimited constituent can be wh-questioned. From
now on, I will refer to this requirement as the ‘wh-questioning’
constraint.
We can easily confirm that the Q-element serving as a delimiter is
always placed in a position where the delimited constituent can be
wh-questioned, even when it is separate from the host wh-phrase. To
take just one example, consider (41):
ð41Þa: Chitra ½ ½kauru aawa ki n ka kataaw d aehuwe?

e e e e e
Chitra who came-A that rumor Q heard-E
Chitra heard the rumor that who came?
b: Chitra mon wa d aehuwe?
e e
Chitra what Q heard-E
What did Chitra hear?
In (41a), the complex DP containing kauru ‘who’ is marked by d . e

The fact that (41b), in which the complex DP is replaced by a simple
wh-phrase, is well-formed indicates that in (41a), d is merged in a
e
position such that the constituent it marks can be wh-questioned.
The example in (42b), by contrast, shows that the clausal constit-
uent TP immediately followed by d in (42a) cannot be wh-questioned:
e
ð42Þa: Chitra ½kauru aawa d kiy la dann wa:

e e e
Chitra who came-A Q that know-A
Chitra knows who came.

b: Chitra ½mon wa d kiy la danne?
e e e
Chitra what Q that know-E
Whati does Chitra know that ti ?
As shown in (42b), the wh-phrase mon wa ‘what’, which is substituted

e
for the clausal complement [kauru aawa] ‘who came’, cannot be a
legitimate wh-constituent. This substitution should be possible if the
constituent marked by d can be wh-questioned.14 Note that (42b) is

e
not ruled out on the basis that the clausal complement of the com-
plementizer requires a finite verb, since (43) is acceptable in some
appropriate context:
ð43Þ Chitra ½kaa- e d kiy la dann wa:

e e e
Chitra who-DAT Q that know-A
Chitra knows to whom.
Example (43) is usable, for instance, in a context in which the speaker is

reporting that Chitra knows the identity of the person to whom Ranjit
gave a book.15 In (43), what precedes the complementizer kiy la ‘that’ e
is a wh-phrase combined with d , indicating that kiy la does not have
e e
to select a constituent containing a verb as its complement.
The impossibility of replacing the clausal constituent with a simple
wh-phrase in (42b) indicates that the ‘wh-questioning’ constraint is
violated by the direct merger of d to its scope position. Thus, the
e
derivation in which d is merged directly to this position is illicit and
e
does not converge in Sinhala. In order to form a legitimate wh-question
where d appears in clause-final scope position, d must first be merged
e e
in a ‘delimiter’ position (for the purpose of identifying a legitimate wh-
constituent), and then moved to the clause-final ‘scope’ position.
In an economy-based approach such as Chomsky (1995b), the
derivation that involves Merge alone, as in (28), is less costly than
the derivation in (27b), which involves both Merge and Move. Thus,
(27b) should, other things being equal, be ruled out in favor of (28).
But in Sinhala wh-questions, the derivation in (28) does not con-
14
This does not mean that natural language never allows the complement clause
that precedes a complementizer to be wh-questioned. Notice that a wh-in-situ lan-
guage like Japanese allows the clausal complement to be turned into a wh-phrase:
ðiÞ John-wa Mary-ni ½nan to itta no?
John-TOP Mary-DAT what that told Q
John told Mary that what?
The unacceptability of (42b), when compared with (i), suggests that while natural
language does allow the option of the clausal complement which precedes a com-
plementizer to be wh-questioned, Sinhala does not choose this option.
15
Example (43) involves sluicing. In the given context, (43) should have a form
like (i) if the entire sequence in the embedded clause is spelled out:
ðiÞ Chitra ½½Ranjit ee pot dunne kaa-
e e d kiy la dann wa:
e e e
Chitra Ranjit that book gave-E who-DAT Q that know-A
Chitra knows to whom it was that Ranjit gave that book.
26 HIDEKI KISHIMOTO
verge because it does not satisfy the ‘wh-questioning’ requirement. In

the economy-based analysis, only convergent derivations are
compared. Thus, the non-convergent derivation in (28) does not
block the derivation in (27b). Therefore, the derivation in (27b),
which involves both Merge and Move, is an optimal derivation and is
selected.16
We can state here that if d , which counts as an operator, is moved to
e
[Spec, C2 ], it forms an operator-variable structure (i.e., an A-bar chain)
with the associated wh-phrase, so that the scope of the wh-phrase is
fixed by the position of d in LF. The present analysis, which analyzes
e
the special verbal marking as assuming the role of a formal feature to
attract a Q-element, can correctly capture the generalization on parti-
cle-predicate concord. That is, when d occurs in Merge position, the
e
scope of the wh-word is marked by the -e marking, which is an overt
realization of a [+Q] feature. But when d is positioned in the scope
e
position, d specifies the scope of the wh-word, and the verb’s special
e
ending does not obtain (since [+Q] is checked off) (cf. section 4).
To summarize, on the assumption that d is construed as a non-
e
projected head, we can naturally explain why d undergoes A-bar
e
operator movement in overt syntax, behaving like a phrasal element,
although it is thought of as being phonologically dependent. I have
argued that d can be merged in any position where the delimited
e
constituent can be turned into a simple wh-expression, and that the
placement of d in clause-final scope position always results from a
e
syntactic operation that moves it after it is first merged in such a
position that it can delimit a wh-constituent.
3. PHRASAL MOVEMENT IN LF
In the previous section, I have argued that when d appears in scope

e
position, it is overtly moved to check the strong [+Q] feature. When
[+Q] is weak, overt movement of d is not required, but for the
e
derivation to converge, [+Q] must be checked in LF.17 In this section,
16
This does not mean that the base-generation of an operator element in scope
position is always prohibited in Sinhala wh-questions. An operator can be merged in
scope position under a very restricted context, and this can happen when a Q-element
cannot serve as a scope marker. I will discuss the details of this fact in section 4.
17
Covert movement cannot be induced by an EPP feature, which may be assigned
to C (cf. Chomsky 2000). If an EPP feature is assigned to C, then overt movement,
rather than covert movement, is induced.
I argue that in such a case, i.e., when [+Q] is weak, d must be moved e
into its scopal position in LF for the purpose of feature checking as
well as forming an operator-variable structure with a wh-phrase, and
that this movement displays the properties of phrasal category
movement.
3.1. Island Effects in LF

As a first point, I show that movement of d must count as phrasal
e
even if it is invoked in LF, and that this covert movement incurs
various island effects in exactly the same way as overt phrasal
movement. Now, for the purpose of checking syntactic islands in
Sinhala, let us consider the following examples that involve overt
pseudo-cleft extraction:
ð44Þa: oyaa ½½Chitra ti dunn pot kieuwe Ranjit- i :

e e e
you Chitra gave book read-E Ranjit-DAT
It was to Ranjiti that you read the book that Chitra gave ti .
b: Chitra ½½Ranjit ti gatta ki n ka kataaw aehuwe
e e e e
Chitra Ranjit bought-A that rumor heard-E
ee pot i :e
that book
It was that booki that Chitra heard the rumor that Ranjit
bought ti .

c: ½Chitra ti kan ko Ranjit pudum unee
e e e maalui :
Chitra ate time Ranjit surprise became-E fish
It was fishi that Ranjit was surprised when Chitra ate ti .
d: ??Chitra ½Ranjit ti gatta kiy la kendiruwe ee pot i :
e e
Chitra Ranjit bought-A that whispered-E that book
It was that booki that Chitra whispered that Ranjit
bought ti .
e: ??Ranjit ½Chitra ti kieuwa d -naedd kiy la danne
e e e
Ranjit Chitra read-A whether that know-E
ee pot i : e
that book
It is that booki that Ranjit knows whether Chitra read ti .
In (44), the focused DPs have been overtly extracted from various
syntactic positions, i.e., from within a relative clause, a noun com-
28 HIDEKI KISHIMOTO
plement clause, an adjunct clause, the complement clause of a non-

bridge verb, and a whether-clause.18 Overt extraction of a phrasal
element via pseudo-clefting is fully acceptable if it does not originate
from within an island, as illustrated in (45):
ð45Þ Ranjit ½Chitra ti kieuwa kiy la kiiwe ee pot i :
e e
Ranjit Chitra read-A that said-E that book
It was that booki that Ranjit said [that Chitra read ti ].
Since the pseudo-cleft construction, which induces the -e suffix on the
verb in the clause where the moved element resides, is a variant of
focusing constructions, I assume that the constituent moved by
pseudo-clefting resides in [Spec, C2 ] to the right, where a focus
operator is typically accommodated in Sinhala (see section 4).19 The
examples in (44) illustrate some of the island effects caused by overt
extraction of phrasal elements.20
Next, let us consider wh-questions. The following wh-questions,
which are the counterparts of those in (44), are all unacceptable in
just the same way as (44), despite the fact that they do not involve
overt extraction:
18
Since d -naedd is a fixed expression, it is treated as an equivalent of ‘whether’ in
e e
English. Even if it turns out that d -naedd is equivalent to ‘whether or not’, the
e e
essential point remains the same, since both types of complement clauses form an
island for extraction:
ðiÞa: ? What do you wonder whether John read?
b: ? What do you wonder whether or not John read?
19
Given this analysis, we can say that in Sinhala, [Spec, C2], which accommodates
a focus operator, stands at the right end of a clause. This means that focus con-
structions in general involve either overt or covert movement of a focusing operator
to [Spec, C2 ] to the right. In this analysis, both clause-final Q-element and overtly
moved ‘focused’ constituent are conceived of as occupying the same [Spec, C2 ], the
only difference being that the Q-element is cliticized onto the verb, behaving like a
clitic, whereas the focused constituent is not.
20
Not all islands are testable due to some language particular restrictions. For
instance, in English, it is known that a factive complement selected by the verb regret
forms a weak island. But in Sinhala, it is not possible to test it, because factive verbs
like kanagaatu wen wa ‘regret’ do not select a clause, but a complex DP, as in (i):
e
ðiÞ Chitra ½Ram tuwaal una

e ki n eka gaena kanagaatu wen wa:
e e e
Chitra Ram hurt became-A that fact about regret become-A
Chitra regrets the fact that Ram was hurt.
A complex DP constitutes an island, independently of the context in which it

appears.
ð46Þa: oyaa ½½Chitra kaa- d dunn pot kieuwe?

e e e e
you Chitra who-DAT Q gave book read-E
To whomi did you read the book that Chitra gave ti ?
b: Chitra ½½Ranjit mon wa d gattae ki n ka kataaw
e e e e e
Chitra Ranjit what Q bought-A that rumor
aehuwe?
heard-E
Whati did Chitra hear the rumor that Ranjit bought ti ?
c: ½Chitra mon wa d kan ko Ranjit pudum unee?
e e e e e
Chitra what Q ate time Ranjit surprise became-E
Whati was Ranjit surprised when Chitra ate ti ?
d: ??Chitra ½Ranjit mon wa d gatta
e kiy la kendiruwe?
e e
Chitra Ranjit what Q bought-A that whispered-E
Whati did Chitra whisper that Ranjit bought ti ?
e: ? Ranjit ½Chitra mon wa d kieuwa d -naedd kiy la
e e e e e
Ranjit Chitra what Q read-A whether that
danne?
know-E
Whati does Ranjit know whether Chitra read ti ?
Recall here that the -e suffix appears on the verb located in the clause
where the wh-phrase acquires scope, but not on any other verbs. The
verb endings in (46) indicate that the sentences are all construed as
direct wh-questions. The island effects in (46) are quite analogous to
those observed in overt pseudo-cleft extraction. This suggests that in
(46), covert phrasal movement of d into [Spec, C2 ] in the main clause
e
should be held responsible for the island effects.
The correctness of this view is supported by the fact that the island
effects are absent in cases where the Q-particle appears to the right of
the islands:
ð47Þa: oyaa ½½Chitra kaa- dunn pot d kieuwe?
e e e e
you Chitra who-DAT gave book Q read-E
You read the book that Chitra gave to who?
b: Chitra ½½Ranjit mon wa gatta eki n e e
Chitra Ranjit what bought-A that
ka kataaw d
e e e aehuwe?
rumor Q heard-E
Chitra heard the rumor that Ranjit bought what?
30 HIDEKI KISHIMOTO
c: ½Chitra mon wa kan ko d Ranjit pudum unee?

e e e e e
Chitra what ate time Q Ranjit surprise became-E
Ranjit was surprised when Chitra ate what?
d: Chitra ½Ranjit mon pot gatta e kiy la d kendiruwe?
e e e
Chitra Ranjit what book bought-A that Q whispered-E
Chitra whispered that Ranjit bought what book?
e: Ranjit ½Chitra mon wa kieuwa d -naedd kiy la d danne?
e e e e e
Ranjit Chitra what read-A whether that Q know-E
Ranjit knows whether Chitra read what?
Recall that the merger of d targets a maximal projection containing
e
a wh-word. Thus, d may be merged to the right edge of an island,
e
separate from its host wh-word, as in (47), if d c-commands and can
e
bind the wh-word. Interestingly, all the examples in (47), in opposi-
tion to those in (46), are acceptable. The only visible difference is that
while the Q-particles are contiguous with their host wh-phrases in
(46), the Q-particles are attached at the right edge of the islands in
(47). The fact that islands effects are voided when Q-particles are
placed at the right edge of islands on the surface was first noted by
Gair (1983).
In all the examples in (46) and (47), the Q-particle d serves as a e
delimiter, and is not moved in overt syntax, so that it must be moved
into its scope position in LF. For the sake of exposition, let us
assume that movement of d proceeds in a successive cyclic fashion
e
from the Merge position to the scope position by way of any
available [Spec, C2] on its way. In all cases in (47), LF movement
starts out at the right edge of the islands, but in (46), it involves
extraction out of the islands. In (46d), d may be raised to its scope
e
position through [Spec, C2] in the embedded clause, and in the rest of
the examples in (46), it is moved directly from the Merge position to
the scope position in LF. Notably, island effects obtain when LF
movement of d involves extraction out of islands, as shown in (46),
e
but they do not when d starts out at the right margin of the islands,
e
as shown in (47).
The presence or absence of the island effects in (46) and (47) (i.e.,
Gair’s generalization) is naturally anticipated in the present analysis,
according to which d is merged to a maximal projection (containing
e
a wh-form) by adjunction:
ð48Þa: ½CP2 ½TP . . . ½XP . . . WH ti . . . . . . d i e
b: ½CP2 ½TP . . . ½XP ½XP . . . WH . . . ti . . . d i e
When d is merged within an island XP, as in (48a), the subsequent

e
LF movement of d into its scope position involves the crossing of
e
XP, as a result of which an island effect is manifested, as illustrated in
(46). But when it is adjoined to XP, as in (48b), it is only dominated
by a single segment of XP. In this case, the later LF movement of d e
to [Spec, C2] would not involve extraction out of XP, so that the
sentences in (47) are all well-formed.21
In this connection, notice that the islandhood of a category
intervening between d and its scopal position cannot be voided by a
e
Merge operation. To illustrate, consider (49):
ð49Þ Chitra ½Ranjit ½½kauru liy pu pot d gatta
e eki n
e e e
Chitra Ranjit who wrote book Q bought-A that
ka kataaw aehuwe?
e e
rumor heard-E
Chitra heard the rumor that Ranjit bought the book that
who wrote?
In (49), d is attached to a DP which comprises a relative clause
e
containing the wh-word kauru ‘who’; this DP is further embedded in a
noun complement clause. The ungrammaticality of (49) is compara-
ble to that of (50), where the lower complex DP is replaced by a
simple wh-phrase:
ð50Þ Chitra ½Ranjit mon wa d gatta e ki n ka kataaw
e e e e e
Chitra Ranjit what Q bought-A that rumor
aehuwe?
heard-E
Whati did Chitra hear the rumor that Ranjit bought ti ?
Both (49) and (50) are matrix wh-questions, as the matrix verb
receives the -e ending. Since d resides in a delimiter position, it must
e
be raised to [Spec, C2] in the main clause in LF. Then, the island
violations in (49) and (50) must be caused by the LF movement of d e
into this scope position. Notice that in (49), d is adjoined to a DP
e
comprising a relative clause, which constitutes an island, represented
by XP in (51):
ð51Þ ½CP2 ½TP . . . ½YP . . . ½XP ½XP . . . WH . . . ti . . . . . . d i
e
21
To be more precise, it is necessary to assume here that no other islands exist
between XP and CP2 .
32 HIDEKI KISHIMOTO
The adjunction operation does not incur an island violation per-

taining to XP. Even so, d must covertly move out of the noun
e
complement island, which is represented by YP in (51). Since the
LF movement of d in (49) crosses the same type of complex DP
e
island as that of (50), (49) is found unacceptable in the same way
that (50) is.
It is important to keep in mind that under the current
assumptions, movement of wh-forms is not necessary throughout
the derivation, since wh-features are [+Interpretable] (see Chomsky
1995b). By the present account, wh-phrases are construed as
variables to be bound by d (without movement) (cf. Pesetsky
e
1987; Reinhart 1997). Thus, as long as a Q-element residing in
scope position appropriately binds the wh-phrase, the scope inter-
pretation can be obtained without requiring movement of the wh-
phrase:
ð52Þ ½CP2 ½TP . . . ½ . . . WHi . . . . . . d i
e
There is a sense in which Sinhala wh-phrases should be construed as

variables to form an A-bar chain with a Q-element, since they are not
inherent wh-operators, but indeterminate pronouns whose quantifi-
cational force is determined according to the kind of Q-particle
associated with them. If wh-words are variables that are not moved
throughout derivations, we can easily see that they can be deeply
embedded within islands when movement of d does not violate e
island conditions, as in (47).
The Sinhala data clearly indicate that LF movement of d incurs e
the same standard island effects that are observed for overt extraction
of phrasal elements. The fact that the island violations observed for
(46) are parallel to those in (44) is expected, given the bare phrase
theoretic assumption on d , since this Q-element can undergo phrasal
e
category movement.
The analysis which views the island violations in (49) as coming
from LF movement of d gains further empirical support from the
e
fact that in (53), where d occurs next to the host wh-phrase, a non-
e
wh-phrase can be extracted without creating any island effect:
ð53Þ Ranjit- i ;
e oyaa ½Chitra ti mon wa d dunne
e e
Ranjit-DAT you Chitra what Q gave-E
kiy la
e dann wa: e
that know-A
To Ranjiti , you know what Chitra gave ti .
In (53), the DP Ranjit- ‘to Ranjit’ is overtly extracted from the

e
embedded clause via scrambling. The acceptability of (53) parallels
the well-formedness of (54), which involves overt extraction of a
phrasal item from a non-interrogative subordinate clause selected by
the verb dann wa ‘know’:
e
ð54Þ Ranjit- i ; oyaa ½Chitra ti ee pot dunna kiy la dann wa:
e e e e
Ranjit-DAT you Chitra that book gave-A that know-A
To Ranjiti , you know that Chitra gave that book ti .
In (53), just like (54), the scrambled phrase can be moved without
displaying any island effect. An embedded wh-question, however,
forms an island for extraction when the Q-element is placed clause-
finally, as in (55):
ð55Þ ??Ranjit- i ; oyaa

e ½Chitra ti mon wa dunna d
e e
Ranjit-DAT you Chitra what gave-A Q
kiy la
e dann wa: e
that know-A
To Ranjiti , you know what Chitra gave ti .
The island effect also obtains when d is attached to the right of the
e
complementizer, as shown in (56):
ð56Þ ? Ranjit- i ; eoyaa ½Chitra ti mon wa dunna kiy la

e e
Ranjit-DAT you Chitra what gave-A that
de danne?
Q know-E
To Ranjiti , what do you know that Chitra gave ti ?
In (56), the matrix verb marks the scope of the wh-phrase, and the Q-
element is base-generated to the right of the complementizer. The facts
in (53), (55) and (56) show that the scrambling of a DP causes an island
violation when d is placed either to the left or to the right of the
e
complementizer, but not when it is contiguous with the wh-phrase.
Now, given that long distance scrambling is an instance of A-bar
movement, as argued by Mahajan (1990), the presence or absence of
the wh-island effects in (53), (55), and (56) can be accounted for on the
assumption that both the Q-particle and the scrambled phrase count
34 HIDEKI KISHIMOTO
as operator elements.22 In (53), d is adjoined to the direct object, and

e
hence does not c-command the indirect object, allowing the extraction
of the DP Ranjit- ‘to Ranjit’ out of the embedded clause with no
e
violation of the Minimal Link Condition (MLC).23 By contrast, in
(55), d appears in [Spec, C2 ] in the complement clause, and in (56), it is
e
adjoined to the right of the complement clause. In both cases, d c- e
commands the indirect object Ranjit- ‘to Ranjit’, so that d counts as a
e e
closer operator element to be attracted. Consequently, the extraction of
Ranjit- to the sentence-initial position via scrambling induces a MLC
e
violation, thereby the sentences being unacceptable.24
22
In Sinhala, scrambling does not allow LF reconstruction, as shown below:

ðiÞ mon wai ; Ranjit ½Chitra ti kieuwa d kiy la dann wa:
e e e e
what Ranjit Chitra read-A Q that know-A
What, Ranjit knows Q Chitra read.
In view of this fact, we can reasonably assume that in Sinhala, a scrambled element
can serve as an operator in LF (cf. Saito 1989, 1992).
23
The MLC is defined as follows (Chomsky 1995b, p. 311):
ðiÞ K attracts a only if there is no b; b closer to K than a; such that K attracts b:

The notion of ‘closeness’ is defined over ‘c-command’ (Chomsky 1995b, p. 358):
ðiiÞ b is closer to the target K than a if b c-commands a: :
The notion of ‘c-command’ is defined as (iii):
ðiiiÞ a c-commands b if a does not dominate b and every c that dominates a
dominates b:
The notion of ‘domination’, which is relevant for ‘c-command’, is defined as follows:
(iv) b is dominated by a if it is dominated by some segment of a.

24
The same contrast that is observed for (53) and (55) obtains in a case like (i),
where the direct object is scrambled across the subject:
ðiÞa: ee pot i ; Ranjit ½kau d ti kieuwe kiy la dann wa:

e e e e
that book Ranjit who Q read-E that know-A
That book, Ranjit knows who read.
b: ??ee pot i ; Ranjit ½kauru ti kieuwa d kiy la dann wa:

e e e e
that book Ranjit who read-A Q that know-A
That book, Ranjit knows who read.
The contrast in acceptability between (ia) and (ib) can be accounted for by the MLC,
which accounts for the contrast in acceptability between (53) and (55).
Notice that the same type of contrast in acceptability that is

observed in (53) and (55) also obtains in the pseudo-cleft construc-
tion, where rightward movement is induced:
ð57Þa: Ram ½Chitra ti mon wa d dunne kiy la danne Ranjit- i

e e e e
Ram Chitra what Q gave-E that know-E Ranjit-DAT
It is to Ranjit that Ram knows what Chitra gave.
b: ??Ram ½Chitra ti mon wa dunna d kiy la
e e e
Ram Chitra what gave-A Q that
danne Ranjit- i : e
know-E Ranjit-DAT
It is to Ranjit that Ram knows what Chitra gave.
Example (57b), which involves the extraction of the dative phrase via
pseudo-clefting, exhibits a mild island effect, but this effect is absent
when d is adjacent to the host wh-phrase, as shown in (57a). The
e
contrast in acceptability can be accounted for by appealing to MLC,
just in the same way as (53) and (55), if the DP undergoing pseudo-
clefting counts as an operator element.25
The non-existence of blocking effects in (53) and (57a) indicates
that when d resides in a delimiter position, no operator (to bind a
e
25
The same type of wh-island violation that is observed for (55) and (57b) occurs
if a phrasal element is extracted from a whether-clause:
ðiÞa: Ranjit ½Chitra ti kieuwa kiy la danne ee pot i :

e e
Ranjit Chitra read-A that know-E that book
It is that book that Ranjit knows that Chitra read.
b: ??Ranjit ½Chitra ti kieuwa d -naedd kiy la danne ee pot i :
e e e e
Ranjit Chitra read-A whether that know-E that book
It is that book that Ranjit knows whether Chitra read.
While long distance extraction is licit from the non-interrogative subordinate clause,
long distance extraction out of the d -naedd -clause is degraded. The deviance of long
e e
distance extraction is also observed when a phrasal item is extracted from the d - e
naedd -clause by way of scrambling:
e
ðiiÞ ??ee pot i ; Ranjit ½Chitra ti kieuwa d -naedd kiy la dann wa:
e e e e e
that book Ranjit Chitra read-A whether that know-A
That book, Ranjit knows whether Chitra read.
In (ib) and (ii), d -naedd ‘whether’ occupies the same position as d , so we can
e e e
assume that d -naedd is an operator element which fills [Spec, C2 ] in the embedded
e e
clause. Under this configuration, the long distance extraction of the DP ee pot ‘that e
book’ will incur a MLC violation.
36 HIDEKI KISHIMOTO
wh-form) occupies [Spec, C2] in overt syntax. Since scope must be

assigned to the wh-phrase by forming an operator-variable structure,
d must be moved to [Spec, C2] in LF if it remains in a delimiter
e
position where it is merged in overt syntax. The data in (53), (55) and
(57) further confirms that the island violations in (46) are indeed
caused by LF movement of d into the scope position. This LF
e
movement is obviously motivated by the LF requirement of scope
interpretations as well as by the necessity of checking the feature
[+Q]. The present analysis is therefore in line with those proposed by
Reinhart (1991) and Longobardi (1992), who argue for the island
sensitivity of LF movement, while maintaining that LF movement
may be necessitated by LF interface conditions. Further, since LF
movement of d displays the same properties as overt phrasal
e
movement, it is easy to see that Q-movement is a phrasal movement,
irrespective of whether it occurs in overt syntax or in LF.
To account for the islands effects found in (46), one might argue
that an abstract operator is generated contiguous with the wh-phrase
and is moved ‘overtly’ to [Spec, C2], adopting Watanabe’s (1992)
analysis. In fact, this line of inquiry is pursued by Gair and
Sumangala (1992) and Sumangala (1992), who argue that when d is e
contiguous with a wh-word, a null pronominal (or a null operator) is
overtly moved to its scope position, which is marked by the verb’s -e
ending. This analysis is not tenable, however. As shown above, when
d appears in delimiter position, no operator element occupies [Spec,
e
C2 ]. It is clear then that we cannot appeal to an overtly moved
abstract operator to account for the island effects.
Notice that in an analysis postulating a null operator, i.e., an
invisible phrasal category, to explain the difference in island effects
between (46) and (47), it is necessary to ensure that the null operator
starts out from the position of its associated Q head. Given the bare
phrase theoretic assumptions, however, even this stipulation is not
necessary, since d can be subject to phrasal movement. In addition,
e
the null operator analysis predicts that the null operator can deter-
mine the scope of a wh-phrase, independently of whether or not d ise
detachable from the host wh-phrase. But as I will discuss in section 4,
scope properties differ depending on whether or not a wh-word is
associated with a detachable Q-element, which suggests that d e
should undergo phrasal movement in (46) and (47). Under the pro-
posed Q-movement analysis, everything falls out at no extra cost, and
there is no need to postulate the existence of a null operator to
account for the island effects in (46).
Hagstrom’s (1998) analysis, which also analyzes Sinhala wh-ques-

tions as invoking Q-movement, differs from my analysis, since it sub-
scribes to the view that d is always generated next to the host wh-phrase,
e
and that when d is embedded in an island, it must migrate to the right
e
edge of the island. Crucially, this ‘migration’ operation does not observe
island conditions; otherwise, wh-phrases would never be embedded in
syntactic islands. From the current perspective, however, there is no
need to stipulate ‘migration’. By my account, since d functions as a wh- e
delimiter in a way comparable to a focus particle like tamay, d may be e
merged in a position separate from its host wh-word. If d is merged in e
such a way that the later movement does not violate any island condi-
tions, as in (47), no island effects are expected to emerge.
At this point, it should also be noted that islandhood can be
voided in wh-questions in which overt movement of Q-elements is
invoked, as shown by the examples in (58):26
ð58Þ a: ½oyaa ½½Chitra kaa- dunn pot kieuwa d kiy la

e e e e e
you Chitra who-DAT gave book read-A Q that
mam dann wa:
e e
I know-A
I know whoi you read the book that Chitra gave to ti .
b: ½Chitra ½½Ranjit mon wa gatta e ki n ka kataaw aehuwa
e e e e
Chitra Ranjit what bought-A that rumor heard-A
d e kiy la mame dann wa:
e e
Q that I know-A
I know whati Chitra heard the rumor that Ranjit bought ti .
c: ½½Chitra mon wa kan ko e Ranjit pudum unaa
e e e
Chitra what ate time Ranjit surprise became-A
d e kiy la mam dann wa:
e e e
Q that I know-A
I know whati Ranjit was surprised when Chitra ate ti .
d: ½Chitra ½Ranjit mon pot gatta e kiy la kendiruwa
e e
Chitra Ranjit what book bought-A that whispered-A
d e kiy la mam dann wa.
e e e
Q that I know-A
I know what booki Chitra whispered that Ranjit bought ti .
26
In all the examples in (58), the intermediate clause is scrambled to the front in
order to facilitate comprehension.
38 HIDEKI KISHIMOTO
e: ½Ranjit ½Chitra mon wa kieuwa d -naedd kiy la aehuwa

e e e e
Ranjit Chitra what read-A whether that asked-A
d e kiy la mam
e dann wa.
e e
Q that I know-A
I know whati Ranjit asked whether Chitra read ti .
In the examples in (58), d is placed to the right of the finite verb in

e
the complement clause selected by the verb dann wa ‘know’, which
e
means that d is overtly moved. But this overt syntactic movement
e
does not induce any island effect. The position where d is originatede
is not overtly marked, but since island conditions pertain to overt
syntactic movement, d must be merged outside the islands.
e
The examples in (58) involve overt movement rather than LF
movement, but still the same strategy for evading island effects is
available. This strategy cannot be used in a language like English,
where wh-words are necessarily moved in the overt component. The
fact indicates that the difference in island effects between ‘wh-in-situ’
languages (like Sinhala) and ‘overt wh-movement’ languages (like
English) is not related to an LF-syntax dichotomy (cf. Huang 1982;
Nishigauchi 1990). Instead, it is reduced to the question of whether or
not a wh-phrase can be associated with a separable Q-element moving
into a scope position. This would be a difference related to the
morphology of wh-words.
To sum up, when d is attached to an island containing a wh-word,
e
as in (47), the LF movement of d does not involve the crossing of the
e
island, so that no island effect obtains.27 However, when d is further e
embedded in another island, an island violation occurs, as illustrated
by (49). These facts clearly indicate that the island violations in (46)
27
One crucial difference that distinguishes Japanese from Sinhala in the voiding
effect of the islandhood is that in Japanese, as reported by Nishigauchi (1990) and
Watanabe (1992), a wh-word cannot be embedded in a whether-clause:
ðiÞ ? John-wa ½Mary-ga nani-o yonda ka-doo-ka sitte-iru no?

John-TOP Mary-NOM what-ACC read whether know Q
What does John know whether Mary read?
If both Sinhala and Japanese freely utilize the strategy to avoid an island effect, the
ungrammaticality of (i) would pose a problem. Although it is beyond the scope of
this paper to discuss the typological difference, the fact seems to suggest that while
d -naedd ‘whether’ in Sinhala does not block the binding of a Q-element to its host
e e
wh-phrase, ka-doo-ka ‘whether’ does if it intervenes between the Q-element and the
host wh-phrase. See Watanabe (1992) and Hagstrom (1998).
are incurred by LF movement of d , which we can assume to be

e
responsible for determining the scope of wh-phrases.
3.2. The Adjunct Condition Effect

One question that remains to be addressed is whether the island
violations that are observed when d is embedded inside islands are
e
incurred by a syntactic operation other than phrasal category
movement, such as feature movement (in the sense of Chomsky
(1995b)) or Agree (in the sense of Chomsky (2000, 2001)). In this
subsection, I argue, in light of the behavior of wh-adjuncts, that
phrasal category movement should be responsible for those island
effects.
One argument in support of this view may be adduced from
Takahashi’s (1997) discussion on the Adjunct Condition on feature
movement.28 Takahashi argues that when phrasal category move-
ment is not involved, i.e., when feature movement is relevant, the
scope of an adjunct is determined via base-generation of an adjunct
operator in the closest operator position (cf. Rizzi 1990). This brings
us the effect that only a short distance construal is allowed for
adjuncts, i.e., the Adjunction Condition effect. In the framework of
Chomsky (2000, 2001), where feature movement is replaced by Agree,
the presence of the Adjunct Condition effect means that the deriva-
tion involving adjuncts does not converge by Agree, and that in order
for the derivation to converge, a null operator must be inserted. But
when a phrasal category is moved, scope determination does not rely
on the insertion of a null operator. In this case, adjuncts are free from
the Adjunct Condition effect, and both short and long distance
construals are allowed.
This means that we can test for the presence or absence of phrasal
category movement by looking at the behavior of adjuncts. In Sinhala,
we can assume that relativization does not invoke phrasal category
movement, since long distance extraction of reason and manner ad-
juncts is not possible, while local construal is perfectly legitimate:
ð59Þa: ½Ranjit ½Chitra ei aawa kiy la kiiw widiy =heetuw
e e e e
i
Ranjit Chitra came-A that said way=reason
the way/reason Ranjit said that Chitra came
28
Takahashi (1997) also argues for feature movement’s susceptibility to the
Subject Condition, but since Sinhala does not show the relevant subject-object
asymmetries, this condition does not apply to Sinhala.
40 HIDEKI KISHIMOTO
b: ½Chitra ei aaw widiy =heetuw

e e e
i
Chitra came way=reason
the way/reason Chitra came
In (59a), the reason and manner adjuncts cannot originate from the
embedded clause, which indicates that the option of inserting a null
operator into [Spec, C2 ] in the local clause must be chosen to form a
relative clause with an adjunct, in the absence of phrasal category
movement.
If phrasal category movement were not implemented for Sinhala
wh-questions, we would predict that long distance dependency should
not be allowed for adjunct wh-phrases. But the manner adjunct
kohom ‘how’ allows for a long distance construal as well as a short
e
distance construal, as shown in (60):
ð60Þa: Ranjit ½Chitra kohom d aawa e e kiy la kiiwe?

e
Ranjit Chitra how Q came-A that said-E
Howi did Ranjit say [that Chitra came ti ]?
b: Chitra kohom d aawe?
e e
Chitra how Q came-E
How did Chitra come?
In (60), LF movement is involved, but long distance dependency is

possible in just the same way as (61), where d is overtly moved to the
e
scope position in the intermediate clause:
ð61Þ ½Ranjit ½Chitra kohom aawa kiy la kiiwa d kiy la
e e e e
Ranjit Chitra how came-A that said-A Q that
mam dann wa:
e e
I know-A
I know howi Ranjit said [that Chitra came ti ].
The well-formedness of (61) is expected in the present analysis, since

the overtly moved Q-element counts as a phrasal category in syntax.
Further, since long distance construal with wh-adjuncts is possible
when phrasal category movement is involved, it must be the case that
phrasal category movement is invoked in the derivation of a wh-
question in (60a), where covert movement is relevant, just as in (61),
where overt movement is relevant. Notice that in the framework of
Chomsky (2000, 2001), no covert movement should exist, but inter-
estingly, the Sinhala data indicate that phrasal category movement

can occur in LF.
In essence, what is demonstrated by Sinhala wh-questions is that
LF movement of d is an instance of phrasal movement which obeys
e
island conditions, and that extraction of d from within an island is
e
illicit even if it occurs in LF. Since Huang (1982), it has been widely
assumed that the bounding conditions are relaxed in the LF com-
ponent, and that LF movement is not subject to island constraints.
But the Sinhala data clearly indicate that LF movement is con-
strained by various island conditions. The voiding effect, however,
can be obtained if a Q-element can be generated in a position where
the later movement does not violate island conditions.
4. NULL OPERATOR INSERTION
Sinhala has a small inventory of wh-items which cannot be associated

with detachable Q-elements. Although this class of wh-phrases has
not received much attention in the Sinhala literature, an inspection of
the data reveals that when the option of Q-movement is not available,
Sinhala resorts to the strategy of null operator insertion to create an
operator-variable structure.
First, the wh-adjunct mok ‘why’ is included in the category of wh-
e
phrases where d can never be separated from the host, although it
e
looks just like an ordinary wh-phrase like kauru ‘who’, which is
associated with a separable Q-element:
ð62Þa: Ranjit ½Chitra mok d aawe e kiy la dann wa:
e e e
Ranjit Chitra why Q came-E that know-A
Ranjit knows why Chitra came.
b: Ranjit ½Chitra mok aawa e d kiy la dann wa:
e e e
Ranjit Chitra why came-A Q that know-A
In the subordinate clause of the verb dann wa ‘know’, the Q-particle
e
d associated with an ordinary wh-phrase can be moved to the clause
e
final position, but in the case of mok , it cannot. There is another type
e
of wh-adjunct, namely, aei ‘why’, which cannot co-occur with d :29 e
29
There are some dialects where the wh-form aei d ‘why’ is permissible. The wh-
e
phrase aei d patterns exactly like mok d ‘why’ in these dialects.
e e e
42 HIDEKI KISHIMOTO
ð63Þ Chitra faei= aei d g pot kieuwe?

e e
Chitra why=why Q book read-E
Why did Chitra read the book?
Since mok and d are never separable, and since the behavior of
e e
mok d is identical to that of aei, as we will see below, it is reasonable
e e
to say that mok d forms a single lexical item, which is unanalyzable
e e
in syntax.
The wh-adjuncts mok d and aei are not associated with a sepa-
e e
rable Q element, which suggests that they cannot invoke movement
of a Q-element to form an operator-variable structure. Now, the
question that immediately arises is how a legitimate operator-variable
format is formed with these wh-adjuncts. There are two conceivable
ways of forming an operator-variable structure. One is to move the
wh-phrases in the LF component. The other is to base-generate a null
operator in the scope position. Note that feature movement and
Agree do not create a licit operator-variable structure encoding scope
for adjuncts, and therefore, are not candidates here (cf. section 3.2).
In the following discussion, I show that Sinhala utilizes the ‘null
operator’ strategy when movement of a Q-element cannot be imple-
mented to assign scope to wh-phrases.
As discussed by Rizzi (1990), and also by Takahashi (1997), the
scope position where an invisible operator can be directly generated is
quite restricted, and it can only be merged in the local scope position
of the clause in which the wh-word resides. But there is no such
stringent ‘locality’ constraint on phrasal category movement. If this is
correct, we can check which option is available by looking at the long
distance construal of wh-questions. Now, consider (64):
ð64Þ ? Ranjit ½Chitra fmok d =aeig aawa kiy la kiiwe?

e e e
Ranjit Chitra why Q=why came-A that said-E
Whyi did Ranjit say [that Chitra came ti ]?
If the wh-adjuncts mok d and aei were susceptible to LF category

e e
raising, they should be able to take matrix scope, but as indicated by
(64), they cannot. In contrast, when the wh-adjuncts take embedded
scope, the sentence is fully acceptable:
ð65Þ Ranjit ½Chitra fmok d =aeig aawe kiy la dann wa:
e e e e
Ranjit Chitra why Q=why came-E that know-A
The fact that the scope of the wh-adjuncts cannot go beyond the
clause in which they are located shows that a null operator is inserted
in the closest [Spec, C2] for scope assignment:
ð66Þ ½CP2 ½TP . . .WHi . . .OPi
In (66), the null operator legitimately binds the wh-word, and thereby
the structure is well-formed. But since a null operator can be merged
only in the local clause where the wh-word is located, the long dis-
tance construal in (67) is not possible:

ð67Þ ½CP2 ½TP . . . ½CP2 ½TP . . . WHi . . .OPi
In (64), the wh-word is not embedded in a syntactic island; this
suggests that the long distance construal would be possible if LF
raising were instantiated. The unacceptability of (64) indicates then
that the scope of the wh-adjuncts is assigned by base-generating a null
operator in the local scope position.
A question worth addressing at this moment is when the null
operator is merged in the local scope position. Since lexical access is
not admitted at LF, and since LF may only involve rearrangement of
constituents, the null operator must be merged in the overt compo-
nent to check a strong [+Q] feature. If this is the case, then it is
predicted that overt extraction of a phrasal element out of the clause
in which a wh-adjunct resides should display a wh-island effect. The
expectation is in fact borne out:

ð68Þ ee pot i ; Ranjit ½Chitra fmok d =aeig ti gatte
e e e
that book Ranjit Chitra why Q=why bought-E
kiy la dann wa:
e e
that know-A
That booki , Ranjit knows why Chitra bought ti .
In (68), a wh-island effect is present, which suggests that an operator
element should occupy the embedded [Spec, C2 ] in overt syntax, in
spite of the fact that it is not visible in the surface strings. The same
island effect is observed for overt pseudo-cleft extraction:

ð69Þ Ranjit ½Chitra fmok d =aeig ti gatte
e e kiy la e
Ranjit Chitra why Q=why bought-E that
danne ee pot i : e
know-E that book
It is that booki that Ranjit knows why Chitra bought ti .
44 HIDEKI KISHIMOTO
The presence of island effects in (68) and (69) stands in sharp contrast
to the lack of a wh-island effect with kohom ‘how’ in (70):
e
ð70Þ ee pot i ;e Ranjit ½Chitra kohom d ti gatte

e e
that book Ranjit Chitra how Q bought-E
kiy la dann wa:
e e
that know-A
That booki , Ranjit knows how Chitra bought ti .
The manner adjunct kohom is associated with a detachable Q-
e
element, and when the Q-element is adjacent to the wh-adjunct, no
operator element occupies the scope position in the embedded clause.
Therefore, (70) is acceptable. Note that a wh-island effect emerges if
d is placed in the clause-final position:
e
ð71Þ ??ee pot i ; Ranjit

e ½Chitra kohom ti gatta
e
that book Ranjit Chitra how bought-E
e
d kiy la dann wa:
e e
Q that know-A
That booki , Ranjit knows how Chitra bought ti .
The fact that the clause in which the wh-adjuncts mok d and aei e e
reside forms an island for long distance extraction of phrasal items,
which is shown in (68) and (69), gives us a clear indication that when
d does not participate in scope determination, the insertion of a null
e
operator into the closest scope position takes place in the overt
component.
In Sinhala wh-questions, the strategy to base-generate a null
operator to form an operator-variable structure is always used when
a detachable d is not available. To see this, consider the following
e
sentences:
ð72Þa: Chitra kiiy-ak d geuwe? e

Chitra how:much Q paid-E
How much did Chitra pay?
b: Chitra kiiya d geuwe? e
Chitra how:much Q paid-E
How much did Chitra pay?
The wh-phrases kiiy-ak ‘how much’ and kiiya ‘how much’ occupy the
same syntactic position, i.e., the direct object position of geuwa ‘paid’,
but differ in the possibility of Q-movement. The wh-phrase kiiy-ak
behaves like an ordinary wh-phrase, in that it permits the associated

Q-element to be placed in clause-final scope position:
ð73Þ Ranjit ½Chitra kiiy-ak geuwa d kiy la dann wa: e e e
Ranjit Chitra how:much paid-A Q that know-A
Ranjit knows how much Chitra paid.
The minimally contrasting wh-form kiiya ‘how much’ behaves dif-
ferently from kiiy-ak ‘how much’:

ð74Þ Ranjit ½Chitra kiiya geuwa d kiy la dann wa:
e e e
Ranjit Chitra how:much paid-A Q that know-A
In (74), d cannot be moved to the end of the embedded clause. In
e
this respect, kiiya ‘how much’ patterns with mok ‘why’.30 e
With this difference in mind, let us see whether or not they permit
the long distance construal of wh-scope. In the case of kiiya ‘how
much’, the matrix scope interpretation is not available when it is
embedded in a subordinate clause:
ð75Þ ? Ranjit ½Chitra kiiya d geuwa kiy la kiiwe?
e e
Ranjit Chitra how:much Q paid-A that said-E
How much did Ranjit say that Chitra paid?
In contradistinction, the scope of kiiy-ak ‘how much’ can readily go

beyond the local clause where it is located:
ð76Þ Ranjit ½Chitra kiiy-ak d geuwa kiy la kiiwe?

e e
Ranjit Chitra how:much Q paid-A that said-E
How much did Ranjit say that Chitra paid?
It goes without saying that the local construals of these two wh-
phrases are allowed. If they take embedded scope, the sentences are
acceptable, as shown in (77):
30
The sole morphological difference between kiiy-ak ‘how much’ and kiiya ‘how
much’ lies in the fact that while the former is accompanied by -ak, the latter is not. In
Sinhala, the same distinction is manifested in the morphology of wh-phrases mok e
‘why’ and mok-ak ‘what’. (The wh-word mok originally means ‘what’, but it is no
e
longer used in this sense.) Interestingly, these wh phrases are similar to kiiya and kiiy-
ak, in that while mok is not associated with a detachable Q-element, mok-ak
e
accompanies a separable Q-element.
46 HIDEKI KISHIMOTO
ð77Þ a: Ranjit ½Chitra kiiy-ak d geuwe kiy la dann wa:

e e e
Ranjit Chitra how:much Q paid-E that know-A
b: Ranjit ½Chitra kiiya d geuwe kiy la dann wa:

e e e
Ranjit Chitra how:much Q paid-E that know-A
Notice that the difference in the possibility of long distance construal

is not determined configurationally, since both wh-phrases occupy the
direct object position of the verb geuwa ‘paid’. Obviously, the dif-
ference should be reduced to the question of whether or not a wh-
phrase is associated with a detachable Q-element.31
The difference in the behavior of the wh-words kiiya and kiiy-ak
is expected in the present analysis. In the first place, with kiiy-ak,
the relevant operator-variable structure can be created by raising d e
to its scope position, so both long and short distance construals
are allowed. In the second, in the case of kiiya, the Q-element
forms part of the lexical item, and the option of raising d is e
not available. In this case, Sinhala has recourse to the strategy of
base-generating a null operator, so only a local construal is permitted.
Interestingly, the long distance construal of relativization is not
blocked from the direct object of the verb geuwa, as indicated by the
well-formedness of (78):
ð78Þ ½Ranjit ½Chitra ei geuwa kiy la kiiw mud lai
e e e
Ranjit Chitra paid-A that said money
the money that Ranjit said that Chitra paid
Example (75) shows that when the Q-particle d originated from the
e
direct object of geuwa cannot be detached, a null operator is inserted
into its local scope position. But (78) suggests that this is a context in
which the deletion of a formal feature is in principle possible without
inserting a null operator. Thus, we can confirm that in wh-ques-
tions, the deletion of [+Q] alone does not yield a legitimate LF rep-
resentation, and that an operator element (to bind a wh-phrase) must
be placed in the scope position. The scope facts observed above can
31
The data presented here suggest that the two types of wh-words cannot be
distinguished by either the ‘argument/adjunct’ distinction (Lasnik and Saito 1984,
inter alia) or the ‘nominal/non-nominal’ distinction (Tsai 1994; Huang 1982), since
using either of these criteria, kiiy-ak ‘how much’ and kiiya ‘how much’ belong to the
same class.
be explained naturally if an operator-variable structure to fix the

scope of a wh-word must be formed by either raising d or inserting a
e
null operator into the scope position (due to LF interface condi-
tions).32
One more question to be answered is why the embedded verb in
(65) retains the special -e ending on the surface even if a null operator
is located in the embedded [Spec, C2]. In the present framework, this
fact may be captured by breaking down the checking operation into
two steps, that is, the checking of a feature and its deletion, as sug-
gested by Chomsky (1995b), and further, by imposing different
requirements on them.
To be more concrete, since the feature [+Q] motivating movement
of d is [-Interpretable], let us assume that the deletion of [+Q] must
e
occur by the LF output. Further, since the strength of features
determines the locus where checking occurs, let us assume that if [+Q]
is strong, it calls for feature checking in overt syntax, but that if [+Q]
is weak, feature checking takes place in LF. Under the current view,
we can say that the checking of a weak [+Q] occurs in LF, because the
derivation does not crash at PF even if a weak feature enters into PF
without checking in overt syntax. But a strong [+Q] feature must be
checked in the overt component; an unchecked strong feature leads to
a crash in the derivation if it feeds into PF (Chomsky 1993).
Suppose further that the deletion of a strong feature [+Q] at the
point of feature checking is preferable to its delayed deletion in LF,
since the latter, but not the former, forces the relevant operations to
occur in two different components. If [+Q] is strong, then both
checking and deletion, in unmarked cases, take place at the same time
in overt syntax. This is what happens when d is overtly moved to the
e
clause-final scope position. However, the deletion of the strong fea-
ture [+Q] is only required for LF convergence by assumption. Thus,
its deletion may be delayed until LF, provided that no convergent
derivation is available when both checking and deletion take place in
overt syntax.
Recall here that the Q-particle d , when it is placed clause-finally,
e
signals the scope of its host wh-phrase, taking over a ‘scope’ marking
function from the special verbal marking (instead of serving as a
delimiter). Since ‘visible’ scope marking indicates the locus where an
32
The ‘null operator’ strategy is not utilized for ordinary wh-words which
accompany a detachable Q-element. I assume that this is due to the fact that these
wh-words do not select null operators in the Numeration.
48 HIDEKI KISHIMOTO
operator acquires scope and is an essential ingredient of scope-

marking constructions, let us assume that the lack of an overt scope
marking on the verb in such constructions leads to a crash on the PF
side. In wh-questions, then, scope must be encoded phonologically
either by the -e ending (i.e., the [+Q] feature) on the verb or by a
clause-final Q-particle affixed to the verb.
Given these assumptions, we can naturally account for the fact
that the verb in (79) must bear the -e ending (even in the presence of a
null operator in [Spec, C2 ]):
ð79Þ ½CP2 ½C0 Chitra aeii pot kieuwe= kieuwa OPi ?
e
Chitra why book read-E=read-A
Why did Chitra read the book?
In (79), since a null operator is merged in overt syntax, the [+Q]

feature on the verb is checked overtly. If [+Q] is deleted at the time of
checking, the verb does not yield the -e marking. But no ‘visible’
elements can take over the ‘scope marking’ function of the -e affix on
the verb, so the deletion of [+Q] in overt syntax leads to a crash on
the PF side. Here, the existence of the -e ending is required in overt
syntax, and therefore, the deletion of [+Q] must be deferred until LF
even if it is checked in the overt component. This explains why the -e
marking is retained on the verb in (79).
This analysis can also explain the fact that in the pseudo-cleft
construction, the verb retains the -e ending even if it involves overt
movement to [Spec, C2 ]:
ð80Þ ½CP2 ½C0 Ranjit ti kieuwe ee pot i

e
Ranjit read-E that book
It was that book that Ranjit read.
The rightward movement of the focused element in (80) can be as-

sumed to be triggered by the strong [+Q] feature. If so, [+Q] is
checked in overt syntax. But if it is deleted in overt syntax, no scope
marker is affixed to the verb (i.e., ee pot ‘that book’ is a full-fledged
e
phrasal constituent, but not a dependent element affixed to the verb),
and the derivation crashes on the PF side. To avoid a PF crash, i.e.,
for the verb to yield a legitimate ‘scope’ marking, the [+Q] feature
must be retained in overt syntax. This analysis explains why the -e
ending obtains in a pseudo-cleft sentence like (80) even if a DP is
overtly moved to [Spec, C2 ]. The present analysis allows the distri-
bution of the -e marking to be characterized optimally.
To summarize, the discussion has shown that a wh-phrase can take

not only a local construal but also a long distance construal when its
associated Q-element can be launched off. When movement of d is e
not an option for generating the scope of wh-in-situ, an invisible
operator is inserted in the local scope position, and in this case, only
the local construal is allowed. In Sinhala wh-questions, an operator-
variable structure must be formed by either raising d or inserting a
e
null operator, in order to assign scope to wh-forms properly. I also
proposed a mechanism to account for the distribution of the special
-e marking with full generality.
5. CONCLUSION
I have argued that with ordinary wh-phrases, Sinhala implements

movement of a Q-element (which counts as a non-projected head) to
fix their scope in wh-interrogatives. A detailed look at Sinhala wh-
questions has shown that movement of a Q-element to form an
operator-variable structure with a wh-form is an instance of phrasal
category movement, which may be induced either in overt syntax or in
LF. The Sinhala data have provided evidence that even when a
Q-particle, which shows the behavior of a phrasal category for the
purpose of movement, is moved in LF, island conditions are observed.
In Sinhala wh-questions, wh-phrases are always assigned scope by
operators external to them. Ordinary wh-phrases are associated with
a Q-element which can serve as a scope assigner, and their scope is
determined by the LF position of the Q-element. For wh-phrases
which do not invoke movement of a Q-element, their scope is
assigned by a null operator which is base-generated in the closest
scope position. The discussion has revealed that in Sinhala, wh-
phrases always serve as variables and are never moved throughout
the derivation, and that even if wh-phrases are deeply embedded in
islands, they can be assigned scope as long as external operators can
legitimately bind them.
REFERENCES
Aoyagi, Hiroshi. 1999. ‘On Association of Quantifier-like Particles with Focus in

Japanese’, in M. Muraki and E. Iwamoto (eds.), Linguistics: In Search of the
Human Mind–A Festschrift for Kazuko Inoue, Tokyo: Kaitakusha, pp. 24–56.
Authier, J.-Marc. 1992. ‘Iterated CPs and Embedded Topicalization’, Linguistic
Inquiry 23, 329–336.
50 HIDEKI KISHIMOTO
Bhatt, Rakesh and James Yoon. 1991. ‘On the Composition of COMP and
Parameters of V2’, Proceedings of the Tenth West Coast Conference on Formal
Linguistics, Stanford: CSLI Publications, pp. 41–52.
Browning, M.A. 1996. ‘CP Recursion and That-t Effects’, Linguistic Inquiry 27, 237–
255.
Choe, Jae W. 1987. ‘LF Movement and Pied-piping’, Linguistic Inquiry 18, 348–353.
Chomsky, Noam. 1993. ‘A Minimalist Program for Linguistic Theory’, in K. Hale
and S. J. Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of
Sylvain Bromberger, Cambridge, MA: MIT Press, pp. 1–52.
Chomsky, Noam. 1995a. ‘Bare Phrase Structure’, in G. Webelhuth (ed.), Government
and Binding Theory and the Minimalist Program, Oxford: Blackwell, pp. 383–439.
Chomsky, Noam. 1995b. The Minimalist Program, Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. ‘Minimalist Inquiries: The Framework’, in R. Martin, D.
Michaels and J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in
Honor of Howard Lasnik, Cambridge, MA: MIT Press, pp. 89–155.
Chomsky, Noam. 2001. ‘Derivation by Phase’, in M. Kenstowicz (ed.), Ken Hale: A
Life in Language, Cambridge, MA: MIT Press, pp. 1–52.
Culicover, Peter. 1992. ‘Topicalization, Inversion, and Complementizers in English’,
ms., Ohio State University.
Gair, James. 1983. ‘Non-configurationality, Movement, and Sinhala Focus’, paper
presented at the Linguistic Association of Great Britain, Newcastle, September
1983. [Published in Gair (1998), pp. 50–64.]
Gair, James. 1998. Studies in South Asian Linguistics: Sinhala and Other South Asian
Languages, Oxford: Oxford University Press.
Gair, James and Lelwala Sumangala. 1992. ‘What to Focus in Sinhala’, Proceedings
of the Eighth Eastern States Conference on Linguistics, Ithaca, NY: CLC Publi-
cations, Cornell University, pp. 93–108.
Hagstrom, Paul. 1998. Decomposing Questions, unpublished Ph.D. dissertation,
MIT.
Huang, C.-T. James. 1982. Logical Relations in Chinese and the Theory of Grammar,
unpublished Ph.D. dissertation, MIT.
Kishimoto, Hideki. 1992. ‘LF Pied Piping: Evidence from Sinhala’, Gengo Kenkyuu
102, 46–87.
Kim, Soo-Won. 1989. ‘The QP Status of Wh-phrases in Korean and Japanese’,
Proceedings of the Eighth West Coast Conference on Formal Linguistics, Stanford:
CSLI Publications, pp. 358–372.
Lasnik, Howard and Mamoru Saito. 1984. ‘On the Nature of Proper Government’,
Linguistic Inquiry 15, 235–289.
Longobardi, Giuseppe. 1992. ‘In Defense of the Correspondence Hypothesis: Island
Effects and Parasitic Constructions in Logical Form’, in C.-T.J. Huang and R.
May (eds.), Logical Structure and Linguistic Structure: Cross-linguistic Perspec-
tives, Dordrecht: Kluwer, pp. 149–196.
Mahajan, Anoop. 1990. The A/A-bar Distinction and Movement Theory, unpublished
Ph.D. dissertation, MIT.
Nishigauchi, Taisuke. 1990. Quantification in the Theory of Grammar, Dordrecht:
Kluwer.
Pesetsky, David. 1987. ‘Wh-in-situ: Movement and Unselective Binding’, in E.J.

Reuland and A.G.B. ter Meulen (eds.), The Representation of (In)definiteness,
Cambridge, MA: MIT Press, pp. 98–129.
Reinhart, Tanya. 1991. ‘Elliptic Conjunction – Non-quantificational LF’, in
A. Kasher (ed.), The Chomsky Turn, Oxford: Blackwell, pp. 360–384.
Reinhart, Tanya. 1997. ‘Wh-in-situ in the Framework of the Minimalist Program’,
Natural Language Semantics 6, 29–56.
Rizzi, Luigi. 1990. Relativized Minimality, Cambridge, MA: MIT Press.
Saito, Mamoru. 1989. ‘Scrambling as Semantically Vacuous A’-movement’, in M.R.
Baltin and A.S. Kroch (eds.), Alternative Conceptions of Phrase Structure, Chi-
cago: University of Chicago Press, pp. 182–200.
Saito, Mamoru. 1992. ‘Long Distance Scrambling in Japanese’, Journal of East Asian
Linguistics 1, 69–118.
Sumangala, Lelwala. 1992. Long Distance Dependencies in Sinhala: The Syntax of
Focus and Wh Questions, unpublished Ph.D. dissertation, Cornell University.
Takahashi, Daiko. 1997. ‘Move-F and Null Operator Movement’, The Linguistic
Review 14, 181–196.
Tsai, Wei-Tien Dylan. 1994. On Economizing the Theory of A-bar Dependencies,
unpublished Ph.D. dissertation, MIT.
Watanabe, Akira. 1992. ‘Subjacency and S-structure Movement of Wh-in-situ’,
Journal of East Asian Linguistics 1, 255–291.
Whitman, John. 1989. ‘Topic, Modality, and IP Structure’, in S. Kuno, I.-H. Lee, J.
Whitman, S.-Y. Bak, Y.-S. Kang and Y.-J. Kim (eds.), Harvard Studies in Korean
Linguistics III: Proceedings of the 1989 Harvard Workshop on Korean Linguistics,
Cambridge, MA: Harvard University, pp. 341–356.
Received 27 October 1997

Revised 17 January 2002
Faculty of Letters
Kobe University
1-1 Rokkodai-cho, Nada-ku
Kobe, 657-8501
Japan
<kishimot@lit.kobe-u.ac.jp>

Kishimoto 2005 Wh-In-Situ and Movement in Sinhala Questions (MLLT 3)

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Kishimoto 2005 Wh-In-Situ and Movement in Sinhala Questions (MLLT 3)

Uploaded by

Copyright:

Available Formats

Natural Language & Linguistic Theory (2005) 23: 1–51 Ó Springer 2005

WH-IN-SITU AND MOVEMENT IN SINHALA QUESTIONSw

ABSTRACT. This article shows that Sinhala, a wh-in-situ language, implements

Sinhala (Indo-Aryan; Sri Lanka) is a wh-in-situ language1 in which

wh-phrase acquires scope, or in a constituent position typically

b: pot i amma ½Siri ti keruwa kiy la kalp naa keruwa:

position where the later movement violates island conditions. When

2.1. Particle-Predicate Concord

ð1Þ Chitra mon wa d gatte?

One characteristic property of questions in which the Q-element d e

ð2Þ Chitra pot gatta:

In a matrix wh-question where the Q-particle d stands next to a wh-

In a wh-in-situ language like Japanese, a question particle occurs only

(4) is ill-formed regardless of whether the verb receives the -e marking

The -e marking must co-occur with the Q particle d located adjacent

ð6Þa: Ranjit ½kau d aawa ekiy la danne? e

In (7), kau d ‘who Q’ is located in the matrix clause but the

A Q-particle d associated with a wh-phrase may occur in clause-

In (10), d appears in the embedded clause, while the host wh-phrase

ð11Þ a: Ranjit ½kau d aawe ekiy la dann wa:

If the Q-element is attached to a wh-phrase, then the -e ending

ðiÞa: Ranjit ½Chitra kiidenek d daekka= daekke kiy la danne?

b: Ranjit ½Chitra kiidenek d daekke kiy la dann wa= danne:

ðiiÞ a: Ranjit ½Chitra kiidenek daekka= daekke kiy la dann wa d ?

b: Ranjit ½Chitra kiidenek daekka d kiy la dann wa= danne:

wh-phrase may not be associated with two instances of the Q-particle

ð12Þa: Ranjit ½kau d aawe=aawa

ð13Þ a: Ranjit ½kauru aawa d kiy la aehuwa:

ðiiiÞa: Ranjit ½Chitra kiidenek d daekke kiy la wiswaas k r n wa=kendiruwa:

b: Ranjit ½Chitra kiidenek daekka d kiy la wiswaas k r n wa=kendiruwa:

ð14Þ Ranjit ½Chitra aawa kiy la dann wa=hiten wa=

As shown in (14), there is no correlation between the class of predi-

ð15Þ a: Q : kiidenek pot kieuwa d ?

By contrast, since wh-attached d is uttered in a context in which the

ð16Þ a: Q: kiidenek d pot kieuwe?

On the other hand, a sentence like (11a) is uttered in a discourse

When a wh-question involving an ordinary wh-phrase like kauru

ordinary direct wh-questions, d cannot be placed clause-ﬁnally.7 The

ð19Þ Chitra ee pot tamay kieuwe:

In (19), the particle tamay speciﬁes a focused constituent, and the -e

ð20Þ Ranjit ee pot kieuwa tamay:

When tamay occurs clause-ﬁnally, the verb takes the ordinary -a

ð21Þa: Chitra ee pot d kieuwe?e e

Unlike the wh-question particle d , the yes/no question particle may

By contrast, the wh-question particle allows for multiple foci:

ðiiiÞ kau d mon pot d kieuwe?

A clause introduced by d -naedd corresponds in form to a whether-

ð23Þ a: Chitra ½mon pot d gatte?

It is not possible to place d immediately after a wh-word embedded

ð24Þ Ranjit ½kauru aawa kiy la d danne= dann wa?

In (24), since an independent scope marking, namely the -e marking

cases, d delimits a focused wh-constituent, and the scope of the wh-

2.2. The Status of Q-Particles

When a Q-particle is merged in its Merge position, it is merged (as an

occurs in both declarative and interrogative clauses, and that a

The lower CP, ﬁlled by an operator such as the Q-element as in (26),

occupies [Spec, C2 ].10 In Sinhala, the operator position [Spec, C2 ] must

I propose that d can potentially be merged in any ‘delimiter’ position

in a wh-question, I assume that its merger is licensed when a focus

In Sinhala, however, there is good reason to believe that d , which is

Example (30a) represents a case in which a wh-word is embedded in a

ð31Þa: Q: Chitra ½ ½kiidenek aawa ki n ka kataaw d aehuwe?